ideaname stringlengths 17 130 | field stringclasses 6
values | subfield stringclasses 30
values | year int64 2.02k 2.03k | url stringlengths 10 259 | pdf_path stringlengths 4 42 | target_prompt stringlengths 392 846 | generation_prompt stringlengths 6.15k 6.6k | yidea stringlengths 11.1k 18.4k | yidea_parsed stringlengths 9.36k 15.9k |
|---|---|---|---|---|---|---|---|---|---|
Positronium laser cooling via the 1³S-2³P transition with a broadband laser pulse | Physics | Molecular Physics | 2,023 | https://arxiv.org/abs/2310.08760 | experiments/backtranslation/papers/33.pdf | We want to perform high-precision spectroscopy on positronium, but the atoms are produced with such a wide range of velocities that Doppler broadening obscures the measurements. Given positronium's extremely short lifetime, standard laser cooling techniques that sequentially address different velocity groups are simply... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to write a research proposal in markdown with three sections: Introduction, Proposal, Discussion. Must address cooling positronium atoms en masse despite short lifetimes and broad velocity distribution. Must propose a new method, plausible, testable, show novelty, etc. Likely idea: use stimulated Ram... | # **Research Proposal**
## 1. Introduction
**Research question** – *How can we remove Doppler broadening from a positronium (Ps) cloud whose atoms are born with a very wide velocity spread, when the available cooling time is limited to the ≈ 140 ns lifetime of ortho‑positronium?*
### Motivation
High‑precision... |
Long-lived valley states in bilayer graphene quantum dots
| Physics | Molecular Physics | 2,024 | https://arxiv.org/abs/2304.00980 | experiments/backtranslation/papers/34.pdf | The valley degree of freedom in bilayer graphene quantum dots is theoretically predicted to be a highly robust basis for quantum information, yet its fundamental relaxation time—the key metric for its stability—has never been experimentally measured. Given that we can already characterize the lifetimes for spin states ... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to write a research proposal in markdown with three sections: Introduction, Proposal, Discussion. Must address measuring valley relaxation time in bilayer graphene quantum dots, compare to spin lifetimes, determine stability.
Need hook. Provide precise idea: Use pulsed gate spectroscopy combined wit... | ## 1. Introduction
**Research question** – *How can we experimentally determine whether the valley degree of freedom in bilayer‑graphene (BLG) quantum dots (QDs) is intrinsically long‑lived enough to serve as a qubit, or whether it relaxes on a timescale that precludes practical use?*
### Motivation
Theory pre... |
Engineering spin coherence in core-shell diamond nanocrystals | Physics | Molecular Physics | 2,023 | https://arxiv.org/abs/2305.03075 | experiments/backtranslation/papers/35.pdf | The practical use of diamond nanocrystals as biological sensors is severely limited because magnetic noise from surface defects drastically shortens the qubit's coherence time, and this problem is worse for the very small particles needed for these applications. Given that fabricating smaller sensors from bulk diamond ... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to write a research proposal, pure Markdown, with sections: Introduction, Proposal, Discussion. Provide hook, novelty, testability etc. Idea: to suppress surface magnetic noise in diamond nanocrystals for NV center qubits.
We need a novel approach. Existing attempts: surface passivation, chemical tr... | ## 1. Introduction
**Research question** – *How can the magnetic noise originating from surface defects of diamond nanocrystals be fundamentally suppressed so that the intrinsic NV‑center spin retains long coherence (T₂) even in particles < 20 nm, without relying on bulk‑diamond micro‑fabrication or simple chemical ... |
Observation of vortices in a dipolar supersolid
| Physics | Molecular Physics | 2,024 | https://arxiv.org/abs/2403.18510 | experiments/backtranslation/papers/36.pdf | Proving the superfluid nature of a dipolar supersolid by observing its rotational response has been inconclusive because its crystal-like structure contributes a rigid-body motion that masks the irrotational flow. Given that any quantized vortices—the definitive sign of superfluidity—would form in the low-density gaps ... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to write a research proposal in markdown with three sections: Introduction, Proposal, Discussion. Must address how to detect quantized vortices in dipolar supersolid where they are hidden in low-density gaps. Provide a concrete experimental method.
Idea: Use phase-sensitive imaging via matter-wave i... | # Research Proposal
## 1. Introduction
### 1.1 Research question
*Can we obtain unequivocal experimental evidence for quantized vortices in a dipolar supersolid, despite the fact that the vortices are expected to reside in the low‑density inter‑droplet channels where standard absorption imaging is blind?*
#... |
Superconductivity in twisted bilayer WSe | Physics | Molecular Physics | 2,024 | https://arxiv.org/abs/2406.03418 | experiments/backtranslation/papers/37.pdf | Moiré heterostructures built from transition metal dichalcogenides are known to host flat electronic bands and various correlated states like Mott insulators, yet unlike in graphene systems, robust superconductivity has consistently failed to appear in experiments. Are we facing a fundamental limitation where the intri... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to produce a research proposal with the described sections. Must be markdown, no LaTeX unless mathematically required; seems conceptual. Provide hook, novelty, etc. Include clarity.
Need to address whether superconductivity is fundamentally suppressed or just missed narrow region. Propose a framewor... | # Research Proposal
## 1. Introduction
**Research question**
Moiré heterostructures of transition‑metal dichalcogenides (TMDs) routinely display narrow, interaction‑driven flat bands and a host of correlated insulating states. Yet, unlike twisted‑bilayer graphene, reproducible superconductivity has not been obs... |
Fault-tolerant quantum computation with a neutral atom processor
| Physics | Molecular Physics | 2,024 | https://arxiv.org/abs/2411.11822 | experiments/backtranslation/papers/38.pdf | Implementing quantum error correction is critical, but the standard approach requires complex syndrome measurement circuits just to diagnose the specific nature of a physical qubit error, like whether it was a bit-flip or a phase-flip. Given that these diagnostic circuits are themselves a major source of computational ... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to produce a research proposal in markdown, with three sections: Introduction, Proposal, Discussion. Must include the required elements: problem framing, hook, precision, derivation, constants/key concepts, novelty, evidence, testability, implementation, impact, feasibility, falsification, scope, ver... | # **Research Proposal**
## 1. Introduction
**Research question**
*How can we reliably detect errors on logical qubits when the very syndrome‑measurement circuits used for error diagnosis are themselves a dominant source of faults and overhead?*
**Motivation**
Current fault‑tolerant architectures rely on **ac... |
(Successful) Democracies Breed Their Own Support | Humanities | Sociology | 2,025 | https://economics.mit.edu/sites/default/files/2023-10/Successful%20Democracies%20Breed%20Their%20Own%20Support.pdf | experiments/backtranslation/papers/159.pdf | While studies show that longer individual exposure to democracy correlates with greater support for it, this simple durational measure fails to explain why this support is often fragile, and we know that broad national indicators like income or education levels are not reliable predictors. Considering that our current ... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to produce a research proposal in markdown, with three sections: Introduction, Proposal, Discussion. Must include problem framing, hook, etc. Should be about disentangling duration of democracy from performance. Provide a framework, perhaps a "Performance-Weighted Democratic Exposure" index, combinin... | # Research Proposal
## 1. Introduction
**Research question**
*How can we separate the simple effect of “years lived under a democratic regime” from the effect of the **tangible performance** (economic growth, corruption, conflict, social welfare) that citizens actually experience during those years?*
**Motiv... |
The Institutional Origins of Shared Prosperity | Humanities | Sociology | 2,024 | https://shapingwork.mit.edu/wp-content/uploads/2025/06/johnson-2025-nobel-lecture-the-institutional-origins-of-shared-prosperity.pdf | experiments/backtranslation/papers/160.pdf | The strong, persistent correlation between a country's economic prosperity and the quality of its institutions, such as protection against expropriation, is well-established, but standard regression analyses are unable to determine the direction of causality. Since it is equally plausible that wealth enables good insti... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to generate a research proposal in markdown, following template: Introduction, Proposal, Discussion. Should include problem framing, hook, precision, derivation, novelty, evidence, testability, implementation, impact, feasibility, falsification, scope, verification, parameter extraction. No math unle... | # Research Proposal
## 1. Introduction
### Research Question
*What is the causal effect of institutional quality—particularly the protection against ex‑propriation—on a country’s long‑run economic development?*
### Motivation
Cross‑sectional regressions consistently show a tight, positive link between toda... |
Learning from Ricardo and Thompson: Machinery and Labor in the Early Industrial Revolution, and in the Age of AI | Humanities | Sociology | 2,025 | https://economics.mit.edu/sites/default/files/2024-04/Learning%20from%20Ricardo%20and%20Thompson%20-%20Machinery%20and%20Labor%20in%20the%20Early%20Industrial%20Revolution%20-%20and%20in%20the%20Age%20of%20AI_0.pdf | experiments/backtranslation/papers/161.pdf | Standard economic models predict that new machinery should raise overall productivity and eventually benefit workers, but the mechanization of the textile trade clearly led to decades of collapsing wages for skilled weavers even as the industry boomed. Since the expected compensatory effects, like the creation of suffi... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to produce a research proposal markdown, with sections Introduction, Proposal, Discussion. Must propose a single coherent framework to explain why mechanization caused collapsing wages, focusing on mechanisms missing from standard models. Provide hook, novelty etc. No math unless inherent; this can b... | # Research Proposal
**Title:** *The Institutional‑Adjustment‑Lag Framework: Why Early Textile Mechanisation Slashed Weavers’ Wages despite Rising Productivity*
---
## 1. Introduction
### Research Question
Standard neoclassical growth models predict that a productivity‑raising technology increases the margina... |
Policy Experimentation in China: The Political Economy of Policy Learning | Humanities | Sociology | 2,024 | http://davidyyang.com/pdfs/experimentation_draft.pdf | experiments/backtranslation/papers/162.pdf | Governments use local pilot programs to learn if new policies will be effective nationally, but this assumes the pilot results are generalizable. In a system where local officials' career advancement depends on demonstrating success, they may be incentivized to select the most promising locations for trials or to apply... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to write a research proposal in markdown, following the template: Introduction, Proposal, Discussion. Must include problem framing, hook, precision, derivation, constants/key concepts, novelty, evidence, testability, implementation, impact, feasibility, falsification, scope, verification, parameter e... | ## 1. Introduction
### Research question
*Governments routinely use locally‑run pilot programs to test whether a new policy will work at the national level. If local officials’ careers depend on “showing success,” they may (i) cherry‑pick the most favourable sites and (ii) devote extraordinary, non‑replicable eff... |
Verbs of perception: A quantitative typological study | Humanities | Linguistics | 2,024 | https://ora.ox.ac.uk/objects/uuid:15d5d68d-f3b6-429c-83bf-d46ef951b34a/files/sq237ht84d | experiments/backtranslation/papers/163.pdf | The theory that a universal, biologically-based hierarchy governs how languages create verbs for the senses rests on evidence from a small, geographically biased sample of languages and doesn't consider other general cognitive or communicative principles known to shape lexicons. If we examine a truly representative glo... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to write a research proposal in markdown following the template. Provide introduction, proposal, discussion. Include problem framing, hook, precision, derivation, novelty, evidence, testability, implementation, impact, feasibility, validation items (falsification, scope, verification, parameter extra... | ## 1. Introduction
### Research question
> *If we examine a truly representative global sample, what patterns of dedicated versus shared verbs for sight, hearing, touch, taste, and smell would actually emerge?*
### Why the question matters
A recent “sensory‑verb hierarchy” hypothesis argues that languages uni... |
The English phrase-as-lemma construction: When a phrase masquerades as a word, people play along | Humanities | Linguistics | 2,024 | https://www.lsadc.org/Files/Language/Language%202025/101.2_03Goldberg%20&%20Shirtz_color.pdf | experiments/backtranslation/papers/164.pdf | Speakers can insert entire phrases into grammatical positions normally occupied by a single word, but attempts to classify these structures as nouns, adjectives, or standard compounds fail to fully account for their unique syntactic properties and distributional patterns. Since these formal classification schemes also ... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to write a research proposal in markdown with three sections: Introduction, Proposal, Discussion. Must include all required elements: problem framing, hook, precision, derivation, key concepts, novelty sentence, evidence, testability, implementation, impact, feasibility, falsification, scope, verific... | # Research Proposal
## 1. Introduction
### Research Question
Speakers regularly insert whole phrases into grammatical positions that, in textbook syntax, are filled by a single lexical item (e.g., *“That **the whole nine‑yards** of paperwork is ridiculous”* or *“She’s a **real piece of work** of a person”*). Co... |
STP: Self-play LLM Theorem Provers with Iterative Conjecturing and Proving
| Computer Science | Artificial Intelligence (AI) | 2,025 | https://arxiv.org/abs/2502.00212 | experiments/backtranslation/papers/87.pdf | The prevailing method for training theorem-proving language models, which involves iteratively fine-tuning on successful proofs, quickly hits a performance ceiling because the remaining unsolved problems in a fixed dataset are too difficult. Given that this iterative process generates almost exclusively failed proofs f... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to write a research proposal in markdown, with three sections: Introduction, Proposal, Discussion. Include hook, problem framing, novelty, testability, etc. It must be about providing a steady stream of useful training data for theorem-proving language models beyond the limits of the initial problem ... | # Research Proposal
## 1. Introduction
### Research Question
*How can we supply a theorem‑proving language model (TPLM) with a continuous, high‑utility training signal once the original benchmark set is saturated, given that the iterative fine‑tuning loop produces almost exclusively failed proof attempts?*
#... |
DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning
| Computer Science | Artificial Intelligence (AI) | 2,025 | https://arxiv.org/abs/2501.12948 | experiments/backtranslation/papers/88.pdf | The standard method for teaching language models complex reasoning requires creating vast, expensive datasets of human-written, step-by-step examples for supervised fine-tuning, which is a major practical and scaling bottleneck. How could a base model be guided to develop these sophisticated reasoning abilities on its ... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to write a research proposal in markdown, structured with three sections: Introduction, Proposal, Discussion. Must include all required components, like problem framing, hook, precision, derivation, constants, novelty, evidence, testability, implementation, impact, feasibility, falsification, scope, ... | ## 1. Introduction
**Research question**
The dominant paradigm for endowing large language models (LLMs) with complex, multi‑step reasoning is *supervised fine‑tuning* on massive collections of human‑written “chain‑of‑thought” (CoT) demonstrations. Constructing such datasets is costly, slow, and does not scale wi... |
Energy-Based Transformers are Scalable Learners and Thinkers
| Computer Science | Artificial Intelligence (AI) | 2,025 | https://arxiv.org/abs/2507.02092 | experiments/backtranslation/papers/89.pdf | Current models that attempt deliberate, multi-step reasoning are either computationally rigid, applying a fixed amount of effort to every problem, or they rely on external supervision like explicit rewards or separate verifier models, which restricts them to specific domains like mathematics and prevents them from lear... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to write a research proposal in markdown, covering intro, proposal, discussion, with many specific subheadings as required. The idea: model that learns to dynamically allocate compute and iteratively refine predictions from unsupervised signal. Propose a self-supervised "Adaptive Computation via Pred... | # Research Proposal
## 1. Introduction
### 1.1 Research Question
*How can we train a single model that (i) decides on‑the‑fly how much computation to spend on a given problem, and (ii) iteratively improves its own answer, using only the raw, unlabeled data that the model is already trained to predict?*
Curr... |
Multiagent Finetuning: Self Improvement with Diverse Reasoning Chains
| Computer Science | Artificial Intelligence (AI) | 2,025 | https://arxiv.org/abs/2501.05707 | experiments/backtranslation/papers/90.pdf | Iteratively finetuning a single language model on its own synthetically generated data is a promising path for self-improvement, but this process consistently fails after a few cycles as the model's outputs become increasingly uniform and cease to offer novel learning signals. Given that any single model, when repeated... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to write a research proposal in markdown, with specific sections. Must include problem framing, hook, precision, derivation, etc. Provide novelty sentence. No math unless needed. Probably propose a framework: "Self-Play with Diversity Amplifiers", using multi-agent ensembles, curriculum generation, a... | # Research Proposal
## 1. Introduction
**Research question**
*Why does iterative self‑finetuning of a single language model on its own synthetically generated data quickly converge to a narrow set of reasoning styles, and how can we construct a perpetually diverse and challenging data stream that continues to i... |
Diffusion forcing: Next-token prediction meets full-sequence diffusion | Computer Science | Artificial Intelligence (AI) | 2,024 | https://arxiv.org/abs/2407.01392 | experiments/backtranslation/papers/91.pdf | Autoregressive models for generating continuous data like video are prone to instability over long horizons due to compounding prediction errors and lack a mechanism for steering generation towards specific goals; conversely, while full-sequence diffusion models are stable and guidable, their inherent design for fixed-... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to write a research proposal per template. Should include problem framing, hook, novelty, etc. Must be 800-1200 words. No equations unless inherently mathematical. The idea: combine autoregressive and diffusion approaches via a "Causal Diffusion Process" that generates variable length sequences by le... | ## 1. Introduction
### 1.1 Research Question
*How can we structure a generative process to produce **stable, guidable, and variable‑length** sequences of continuous data (e.g., video, motion capture, audio)?*
### 1.2 Motivation
Autoregressive (AR) models excel at **causal** generation: each new frame is condi... |
Product of Experts for Visual Generation
| Computer Science | Artificial Intelligence (AI) | 2,025 | https://arxiv.org/abs/2506.08894 | experiments/backtranslation/papers/92.pdf | We have powerful, pre-trained generative models for visual realism, separate models for semantic understanding, and even non-neural sources like physics simulators, but retraining a single, monolithic system to unify this diverse knowledge is impractical. Since generating candidates and then filtering them is computati... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to craft a research proposal in markdown, covering introduction, proposal, discussion, meeting all the listed requirements. The problem: integrating pre-trained visual generative models, semantic models, physics simulators etc., without retraining a monolithic system, and avoiding generate-then-filte... | ## 1. Introduction
**Research question** – *How can we fuse heterogeneous sources of knowledge (high‑fidelity visual generators, semantic understanding modules, and physics simulators) *during* the generative process, without resorting to expensive generate‑then‑filter pipelines or costly joint retraining?*
### M... |
Optimality of Gerver's Sofa
| Mathematics | Geometry | 2,024 | https://arxiv.org/abs/2411.19826 | experiments/backtranslation/papers/57.pdf | The best known shapes that can navigate a right-angled corridor are constructed by piecing together simple geometric elements like squares and circular arcs, but this approach seems unlikely to yield the true maximum area. How can we systematically determine the boundary of a shape that maximizes its area by continuous... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to produce a research proposal in markdown. The question: "The best known shapes that can navigate a right-angled corridor are constructed by piecing together simple geometric elements like squares and circular arcs, but this approach seems unlikely to yield the true maximum area. How can we systemat... | ## 1. Introduction
**Research question**
*What shape of maximal area can be carried through a right‑angled corridor while always touching both the inner (corner) wall and the outer wall?*
The classic “square + quarter‑circle” construction demonstrates that a shape built from a few elementary pieces can pass the... |
Stability of Tori under Lower Sectional Curvature
| Mathematics | Geometry | 2,023 | https://arxiv.org/abs/2307.03824 | experiments/backtranslation/papers/58.pdf | When a sequence of Riemannian manifolds collapses under a lower sectional curvature bound, powerful fibration theorems can describe the topology of the limit space, but these theorems fundamentally require the limit to be a smooth manifold. If every manifold in the sequence is known to be a torus, the limit space can s... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to write a research proposal in markdown with three parts: Introduction, Proposal, Discussion. Must address problem: sequences of Riemannian manifolds collapsing under lower sectional curvature bound, fibration theorems need smooth limit; we have all manifolds are tori, limit may be singular; need to... | # Research Proposal
## 1. Introduction
**Research question.**
*When a sequence of Riemannian \(n\)-tori \(\{(T^{n},g_{i})\}_{i\in\mathbb N}\) collapses under a uniform lower sectional‑curvature bound \(\sec_{g_{i}}\ge K_{0}\) and the Gromov–Hausdorff limit \(X\) is a highly singular compact length space, does t... |
Tropical intersection homology
| Mathematics | Geometry | 2,024 | https://arxiv.org/abs/2412.20748 | experiments/backtranslation/papers/59.pdf | For smooth toric varieties, tropical cohomology provides a direct geometric interpretation of the Chow group modulo numerical equivalence, but this framework breaks down for general smooth proper varieties because their tropicalizations are singular. Since this singularity causes the standard tropical cohomology to fai... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to write a research proposal with three sections: Introduction, Proposal, Discussion. Must be pure Markdown, include LaTeX if necessary (it's a mathematical problem). Must address constructing a well-behaved cohomology theory on singular tropical spaces that recovers intersection numbers.
We must in... | ## 1. Introduction
### 1.1 Research question
For a smooth proper variety \(X\) over a non‑Archimedean field, the tropicalisation \(\operatorname{Trop}(X)\) is a polyhedral space equipped with integer weights on its top‑dimensional cells. When \(\operatorname{Trop}(X)\) happens to be a smooth fan (the case of a s... |
Synthetic continued pretraining
| Computer Science | Artificial Intelligence (AI) | 2,024 | https://arxiv.org/abs/2409.07431 | experiments/backtranslation/papers/60.pdf | Pretrained language models struggle to parametrically learn new knowledge from small, specialized corpora because standard continued pretraining is highly data-inefficient, requiring facts to be presented in numerous diverse contexts. Since simply paraphrasing the source documents fails to generate this necessary repre... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to produce a research proposal in markdown, following the specified template: Introduction, Proposal, Discussion. Must include all required sections: problem framing, hook, precision, derivation, constants, novelty, evidence, testability, implementation, impact, feasibility, falsification, scope, ver... | # Research Proposal
## 1. Introduction
### Research Question
*How can we embed the factual content of a **small, specialized corpus** into the weights of a pretrained language model (LM) without the data‑inefficiency of standard continued pre‑training and without relying on naïve paraphrasing that fails to crea... |
Mixture of Parrots: Experts improve memorization more than reasoning
| Computer Science | Artificial Intelligence (AI) | 2,024 | https://arxiv.org/abs/2410.19034 | experiments/backtranslation/papers/61.pdf | The Mixture-of-Experts architecture allows us to drastically increase the total number of parameters in a model with minimal additional computational cost, but it is not well understood what kind of performance is gained by adding more experts versus simply making the dense parts of the model larger. Given that only a ... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to write a research proposal in markdown, following template. Must include introduction, proposal, discussion. Need to address the given question: comparing scaling number of experts (inactive experts) vs scaling active model width for tasks requiring retrieval of stored facts vs multi-step reasoning... | # Research Proposal
**Title:** *Active‑vs‑Inactive Capacity in Mixture‑of‑Experts: When Does Adding “Silent” Experts Beat Widening the Active Network?*
**Authors:** *[Your Name]*, *[Collaborators]*
**Institution:** *[Affiliation]*
---
## 1. Introduction
### 1.1 Research Question
> **How do the perfor... |
Repeat After Me: Transformers are Better than State Space Models at Copying | Computer Science | Artificial Intelligence (AI) | 2,024 | https://arxiv.org/abs/2402.01032 | experiments/backtranslation/papers/62.pdf | New sequence models that compress the entire input history into a fixed-size state are achieving impressive performance on standard benchmarks, yet this architectural constraint creates a fundamental information bottleneck. Given that common benchmarks may not be sensitive to this limitation, how can we isolate and mea... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to write a research proposal in markdown with three sections: Introduction, Proposal, Discussion. Must address isolating and measuring performance consequences of fixed-size state bottleneck in sequence models. Provide problem framing, hook, precision, derivation, concepts, novelty, evidence, testabi... | ## 1. Introduction
**Research question**
> *New sequence models that compress the entire input history into a fixed‑size state are achieving impressive performance on standard benchmarks, yet this architectural constraint creates a fundamental information bottleneck. Given that common benchmarks may not be sensiti... |
Constable: Improving Performance and Power Efficiency by Safely Eliminating Load Instruction Execution
| Computer Science | Computer Systems | 2,024 | https://arxiv.org/abs/2406.18786 | experiments/backtranslation/papers/111.pdf | Techniques like load value prediction and memory renaming are effective at breaking data dependencies, but they do not solve the problem of resource dependencies because the load instruction must still execute to verify the prediction, consuming critical hardware like reservation station entries and cache ports. Given ... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to write a research proposal in markdown, with Introduction, Proposal, Discussion. Must address the problem: load value prediction and memory renaming break data dependencies but still have resource dependencies because load must execute to verify prediction, consuming reservation station entries, ca... | # Research Proposal
## 1. Introduction
### 1.1 Research Question
> **How can we design a micro‑architectural mechanism that mitigates the performance impact of *both* data dependencies and resource dependencies caused by load instructions?**
Current techniques—load‑value prediction (LVP) and memory renaming... |
Measuring Internet Routing from the Most Valuable Points
| Computer Science | Computer Systems | 2,024 | https://arxiv.org/abs/2405.13172 | experiments/backtranslation/papers/112.pdf | Analyzing the complete set of internet routing data from public repositories is becoming computationally infeasible due to its quadratic growth, forcing researchers to rely on ad-hoc sampling methods like randomly choosing a subset of monitoring points. Since this arbitrary selection provides no guarantee of informatio... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to produce a research proposal in markdown with three parts: Introduction, Proposal, Discussion. Must present idea for selecting compact subset of sources (monitoring points) to maximize observational coverage. Should be principled, avoid random sampling. Provide hook, novelty, testability, etc.
We ... | ## 1. Introduction
### Research Question
Public BGP collectors (e.g., RouteViews, RIPE RIS) now produce **quadratically growing** streams of routing updates. Storing and analysing the *complete* dataset is becoming infeasible, forcing researchers to **sample** a subset of monitoring points (sources). Randomly pi... |
Minder: Faulty Machine Detection for Large-scale Distributed Model Training
| Computer Science | Computer Systems | 2,024 | https://arxiv.org/abs/2411.01791 | experiments/backtranslation/papers/113.pdf | In large-scale distributed training, manually inspecting logs to find a single faulty machine among thousands is too slow, yet automated approaches fail because no single monitoring metric reliably indicates a fault and the definition of a "normal" baseline changes with every unique training task. Given that simple thr... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to write a research proposal with the given structure. Must be in pure Markdown.
We need to propose a system to automatically and accurately identify a specific faulty machine among thousands during large-scale distributed training, without simple thresholds or supervised methods.
Idea: Use online ... | ## 1. Introduction
**Research question**
*In large‑scale distributed deep‑learning training, how can we automatically and accurately pinpoint the single physical machine that is causing a failure, **in real time**, when (i) no individual monitoring metric reliably signals a fault, and (ii) the notion of “normal” c... |
PowerInfer: Fast Large Language Model Serving with a Consumer-grade GPU
| Computer Science | Computer Systems | 2,024 | https://arxiv.org/abs/2312.12456 | experiments/backtranslation/papers/114.pdf | To run a large language model on a personal computer, we must offload entire model layers from the small, fast GPU memory to the large, slow CPU memory, but this approach is fundamentally inefficient because every single token generated requires accessing all layers, creating a severe bottleneck over the slow hardware ... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to write a research proposal. Should not use LaTeX unless needed. The idea: decide which parts of model to keep on GPU vs CPU to minimize latency. Propose a profiling-driven, granularity-aware partitioning based on "criticality" and "memory bandwidth sensitivity". Perhaps a two-stage algorithm: (1) c... | # Research Proposal
## 1. Introduction
### Research Question
*How can we decide which specific parts of a large language model (LLM) should reside on a fast GPU versus a large, slow CPU memory so that the end‑to‑end latency of token generation is minimized?*
### Motivation
Running a modern LLM (e.g., 70 B‑... |
NanoFlow: Towards Optimal Large Language Model Serving Throughput
| Computer Science | Computer Systems | 2,024 | https://arxiv.org/abs/2408.12757 | experiments/backtranslation/papers/115.pdf | Despite our understanding that large language model serving is primarily limited by compute, existing systems fail to keep GPUs fully utilized because they execute the different stages of inference sequentially. Since compute-intensive, memory-intensive, and network-intensive operations are run one after another, the G... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to write a research proposal in markdown with three sections: Introduction, Proposal, Discussion. Must include all required elements: problem framing, hook, precision, derivation, constants/key concepts, novelty, evidence, testability, implementation, impact, feasibility, falsification, scope, verifi... | # Research Proposal
## 1. Introduction
### 1.1 Research Question
*How can we serve large language models (LLMs) without the strict sequential execution of compute‑intensive, memory‑intensive, and network‑intensive stages, thereby keeping GPUs fully utilized?*
### 1.2 Motivation
Current LLM serving stacks... |
Error Credits: Resourceful Reasoning about Error Bounds for Higher-Order Probabilistic Programs | Computer Science | Programming Languages (PL) | 2,024 | https://arxiv.org/abs/2404.14223 | experiments/backtranslation/papers/116.pdf | Formal verification of probabilistic programs is hampered by the fact that our logics treat error probabilities as fixed annotations on judgments, forcing us to use crude union bounds for composition which yields overly pessimistic results. This structure also fundamentally prevents us from specifying how a function's ... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to produce a research proposal in markdown, meeting the requirements. The core idea: treat error probabilities as value-dependent, using a probabilistic Hoare logic with dependent types or symbolic probability expressions, possibly using “probability polynomials” and a “probability assertion language... | # **Research Proposal**
## 1. Introduction
**Research Question**
*How can we reason about error probabilities in probabilistic programs so that the bound on a function’s failure chance can depend on the actual values of its inputs, rather than being a fixed annotation that forces overly conservative union‑bound c... |
Tour de gross: A modular quantum computer based on bivariate bicycle codes | Physics | Quantum Physics and Computing | 2,025 | https://arxiv.org/abs/2506.03094 | experiments/backtranslation/papers/1.pdf | Quantum low-density parity-check codes promise a significant reduction in the number of physical qubits required for fault tolerance compared to surface codes, but they are widely considered impractical because they demand complex, long-range connections between qubits. Given that our physical hardware is fundamentally... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to write a research proposal addressing how to design a practical fault-tolerant architecture that realizes efficiency of quantum LDPC codes with only local 2D connectivity. Must be in markdown, 800-1200 words, follow template: Introduction, Proposal, Discussion. Must include problem framing, hook, p... | ## 1. Introduction
**Research question**
*How can we realise the qubit‑efficiency of quantum low‑density parity‑check (LDPC) codes on hardware that is essentially limited to two‑dimensional, nearest‑neighbour (NN) connectivity?*
**Why it matters**
Surface‑code architectures dominate today’s fault‑tolerant roa... |
Roadmap to fault tolerant quantum computation using topological qubit arrays | Physics | Quantum Physics and Computing | 2,025 | https://arxiv.org/abs/2502.12252 | experiments/backtranslation/papers/2.pdf | The theory of topological quantum computing suggests that qubits based on Majorana modes could be inherently robust and naturally suited for error correction through direct multi-qubit measurements, unlike conventional platforms that build these from complex gate sequences. However, without a concrete, step-by-step exp... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to produce a research proposal with the described components. Must be in Markdown, no equations unless inherently mathematical; likely non-mathematical. Provide incremental hardware goals and validation protocols for scaling Majorana-based topological quantum computing. Follow template: Introduction,... | ## 1. Introduction
### Research question
*How can we turn the promise of intrinsically protected Majorana qubits into a concrete, step‑by‑step experimental roadmap that leads from a single reliable qubit to a small processor capable of demonstrating quantum error‑detection through direct multi‑qubit parity measure... |
Kramers nodal lines in intercalated TaS2 superconductors | Physics | Quantum Physics and Computing | 2,025 | https://arxiv.org/abs/2503.08570 | experiments/backtranslation/papers/3.pdf | The theoretical framework for Kramers nodal lines in noncentrosymmetric achiral crystals is established, yet all existing material candidates have proven unsuitable for clear experimental verification because they either possess complex electronic structures with multiple bands crowding the Fermi level or their nodal l... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to write a research proposal with the given structure. The idea: identify or engineer a material with isolated Kramers nodal line at Fermi level. Approach: combine high-throughput computational screening with symmetry-guided design and strain/chemical substitution tuning, plus interface engineering (... | ## 1. Introduction
**Research question**
*How can we identify or engineer a real, non‑centrosymmetric achiral crystal that hosts **one** isolated Kramers nodal line (KNL) exactly at the Fermi level, so that the line can be observed unambiguously in experiment?*
**Why the problem matters**
The Kramers nodal‑li... |
Detecting single gravitons with quantum sensing | Physics | Quantum Physics and Computing | 2,023 | https://arxiv.org/abs/2308.15440 | experiments/backtranslation/papers/4.pdf | The calculated rate for an atom to absorb a single graviton is so infinitesimally small that it is considered unobservable, and our large-scale interferometers can only register the classical strain from waves composed of innumerable gravitons. Given that these existing approaches at both the microscopic and macroscopi... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to write a proposal about detecting a single graviton energy deposition. Must be in markdown, ~800-1200 words. Must follow three-part format. Should include problem framing, hook, precision, derivation, novelty, evidence, testability, implementation, impact, feasibility, falsification, scope, verific... | ## 1. Introduction
**Research question**
*Can we devise a laboratory experiment that is capable of registering the deposition of a *single* quantum of gravitational‑wave energy (a graviton) into a material system?*
**Why this matters**
The standard detection strategies for gravitational radiation occupy two d... |
Geometry-Information Duality: Quantum Entanglement Contributions to Gravitational Dynamics
| Physics | Quantum Physics and Computing | 2,024 | https://arxiv.org/abs/2409.12206 | experiments/backtranslation/papers/5.pdf | General relativity's field equations define spacetime curvature as a response to the distribution of energy and momentum, and even semi-classical approaches are limited to using the local expectation value of the quantum stress-energy tensor as the source. This framework appears to neglect any direct gravitational role... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to produce a research proposal in markdown, following the three-part template: Introduction, Proposal, Discussion. Must include problem framing, hook, precision, derivation, constants, novelty, evidence, testability, implementation, impact, feasibility, falsification, scope, verification, parameter e... | ## 1. Introduction
**Research question** – *How can we formulate a dynamical principle for gravity in which the spacetime metric is directly sourced by the non‑local entanglement structure of a quantum field, rather than solely by the local expectation value ⟨T\_{\mu\nu}⟩?*
### Motivating gap
General Relativit... |
Geometrically Enhanced Topological Quantum Codes | Physics | Quantum Physics and Computing | 2,025 | https://arxiv.org/abs/2505.10403 | experiments/backtranslation/papers/6.pdf | While rotating the lattice of a two-dimensional toric code is a known technique for reducing the number of physical qubits required, this geometric optimization has not been generalized, leaving higher-dimensional topological codes constructed on simple hypercubic lattices with very high resource costs. How can we deve... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to produce a research proposal in markdown with three sections: Introduction, Proposal, Discussion. Must include problem framing, hook, etc. The idea: systematic method for alternative lattice geometries in higher dimensions for topological codes with fewer qubits. We can propose a framework: "Lattic... | # Research Proposal
## 1. Introduction
### 1.1 Research Question
*How can we develop a systematic method for defining and analysing alternative lattice geometries in higher dimensions so that topological stabiliser codes require **significantly fewer physical qubits** for a given level of error protection?*
... |
How to factor 2048 bit RSA integers with less than a million noisy qubits
| Physics | Quantum Physics and Computing | 2,025 | https://arxiv.org/abs/2505.15917 | experiments/backtranslation/papers/7.pdf | A recently proposed method using approximate residue arithmetic breaks the long-held assumption that quantum factoring requires arithmetic registers as large as the number being factored, offering a path to significant qubit reduction. This new approach, however, is practically unusable as it introduces a thousand-fold... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to produce a research proposal in markdown, following three-part format exactly: Introduction, Proposal, Discussion. Must include problem framing, hook, precision, derivation, constants/key concepts, novelty, evidence, testability, implementation, impact, feasibility, falsification, scope, verificati... | ## 1. Introduction
**Research question** – *How can we retain the dramatic qubit savings offered by approximate‑residue arithmetic for Shor’s modular exponentiation while removing the prohibitive thousand‑fold increase in gate operations?*
### Motivation
Factoring a composite integer \(N\) with Shor’s algorith... |
The Apocalypse from Below: The Dangerous Idea of the End of the World, the Politics of the Oppressed, and Anti-Anti-Apocalypticism | Humanities | Political theory | 2,024 | https://www.cambridge.org/core/services/aop-cambridge-core/content/view/054311423D0D7BC9E3B7586C894F0CAB/S0003055424000479a.pdf/the-apocalypse-from-below-the-dangerous-idea-of-the-end-of-the-world-the-politics-of-the-oppressed-and-anti-anti-apocalypticism.pdf | experiments/backtranslation/papers/141.pdf | Political theory largely dismisses apocalyptic rhetoric as inherently dangerous and authoritarian because it seems to impose a singular, unchangeable future, and recent defenses focusing on a movement's conduct or intended goals do not resolve this core problem. Given that such rhetoric is deployed by both powerful eli... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>I need to produce a research proposal in markdown with three sections: Introduction, Proposal, Discussion. The content: need to propose a single coherent framework to evaluate apocalyptic rhetoric legitimacy beyond presentation or goals, establishing a more fundamental principle. Should be precise, testable,... | # Research Proposal
## 1. Introduction
### 1.1 Research Question
*How can we formulate a principled, content‑independent standard for judging the legitimacy of apocalyptic rhetoric (AR) that is applicable to both powerful elites and structurally marginalized groups?*
Current political‑theoretic accounts tre... |
AI, Global Governance, and Digital Sovereignty
| Humanities | Political theory | 2,024 | https://arxiv.org/abs/2410.17481 | experiments/backtranslation/papers/142.pdf | Current analyses of global power dynamics struggle to make sense of the rise of artificial intelligence, where states are pursuing national sovereignty over technologies that are almost exclusively developed and controlled by private corporations. Given that our established international relations frameworks tend to tr... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to write a research proposal with sections Introduction, Proposal, Discussion, in markdown. Must include problem framing, hook, precision, etc. Provide a single coherent framework to chart interplay of state and corporate power in AI domain. Should be non-mathematical? Could include conceptual framew... | # Research Proposal
## 1. Introduction
### 1.1 Research Question
*How can we systematically chart the intertwined exercise of public (state) and private (corporate) authority over artificial intelligence (AI) when influence is exercised not only through direct coercion or market transactions but also by reshapi... |
A Products liability Framework for AI | Humanities | Law | 2,024 | https://journals.library.columbia.edu/index.php/stlr/article/download/12763/6287/34073 | experiments/backtranslation/papers/143.pdf | The legal concept of product liability hinges on identifying a defect that exists at the time a product is sold, and our regulatory models are built around approving a fixed, finished design before it reaches the market. Since complex AI systems can now learn and change their own operational logic long after they are d... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to produce a research proposal in markdown, with three sections: Introduction, Proposal, Discussion, following the guidelines. Include problem framing, hook, precision etc.
The topic: legal product liability for adaptive AI systems that change after deployment. Need a framework that allows accountab... | # Research Proposal
**Title:** *Dynamic Accountability Ledger (DAL): A Trace‑Based Legal Framework for Product‑Liability of Continually‑Learning AI Systems*
---
## 1. Introduction
### 1.1 Research Question
*How can product‑liability law hold parties accountable for harms caused by AI systems that autonomou... |
How Generative AI Turns Copyright Upside Down
| Humanities | Law | 2,024 | https://law.stanford.edu/wp-content/uploads/2024/09/2024-09-30_How-Gerative-AI-Turns-Copyright-Upside-Down.pdf | experiments/backtranslation/papers/144.pdf | Copyright law is built on two core principles: we protect the specific expression of a work rather than the general idea, and we rely on the similarity of that expression to prove infringement. Given that new generative systems automate the creation of detailed expression based on a user's simple prompt, and that two u... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to draft a research proposal with the given structure. Must propose a single coherent framework for determining protectability and copying in generative AI context. Should be non-mathematical, but could include some conceptual metrics. Need hook, precision, derivation, novelty, testability, etc. Prov... | # Research Proposal
**Title:** *Prompt‑Expression Attribution Framework (PEAF): A Structured Legal‑Technical Test for Protectability and Copying in Generative‑AI Works*
---
## 1. Introduction
### Research Question
> *How can the classic copyright doctrines of “idea‑expression dichotomy” and “similarity‑based... |
Public Trust Principles and Environmental Rights: The Hidden Duality of Climate Advocacy and the Atmospheric Trust | Humanities | Law | 2,025 | https://journals.law.harvard.edu/elr/wp-content/uploads/sites/79/2025/03/04_HLE_49_1_Ryan.pdf | experiments/backtranslation/papers/145.pdf | Legal attempts to compel government action on climate change by expanding the common law public trust doctrine to the atmosphere are frequently dismissed by courts as exceeding the doctrine's traditional scope, while separate arguments for a fundamental right to a stable climate are hampered by the fact that American c... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to write a research proposal addressing the legal question: constructing a legally cognizable claim to hold government accountable for protecting climate system, given limits of public trust doctrine and fundamental rights. Need to propose a new legal framework—maybe a "Climate Fiduciary Duty" ground... | ## 1. Introduction
### Research Question
*How can a legally cognizable claim be fashioned that obliges the U.S. government to protect the climate system, given that the classic public‑trust doctrine (as applied to the atmosphere) is routinely deemed beyond its traditional scope and that constitutional “fundamental... |
Ecocentric Environmental Justice: Why We Should Go There and How We Can Get There | Humanities | Law | 2,024 | https://digitalcommons.pace.edu/cgi/viewcontent.cgi?article=1913&context=pelr | experiments/backtranslation/papers/146.pdf | The environmental justice movement successfully uses a human-centered framework to address the disproportionate environmental burdens on marginalized communities, yet this approach is inherently limited because it only recognizes harm to people. Since our legal and ethical tools are designed to protect human interests,... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to produce a research proposal in markdown with three parts: Introduction, Proposal, Discussion. Must include all required sections: problem framing, concise hook, precision, derivation, constants/key concepts, novelty, evidence, testability, implementation, impact, feasibility, falsification, scope,... | ## 1. Introduction
**Research Question**
*How can we create a more holistic approach to environmental protection when the dominant justice framework separates human communities from the natural world they inhabit?*
**Motivation**
The environmental‑justice (EJ) movement has demonstrated that centering the live... |
Scaling Laws for Precision
| Computer Science | Artificial Intelligence (AI) | 2,024 | https://arxiv.org/abs/2411.04330 | experiments/backtranslation/papers/69.pdf | Established scaling laws effectively guide the trade-off between model and data size, but they are fundamentally incomplete as they only apply to high-precision arithmetic and cannot explain phenomena like performance degradation from more training when models are later quantized. Since using lower precision is critica... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to output markdown, following three-part format: Introduction, Proposal, Discussion. Need to address problem: scaling laws missing precision effects, need unified predictive model for interplay between precision, model parameters, dataset size. Provide concise hook, novelty statement, testability, et... | ## 1. Introduction
**Research question** – *How can we build a predictive framework that simultaneously accounts for model size, dataset size, and numerical precision, so that we can anticipate both the gains of larger models/train‑sets *and* the performance loss that appears when those models are later quantized?* ... |
KAN: Kolmogorov-Arnold Networks
| Computer Science | Artificial Intelligence (AI) | 2,024 | https://arxiv.org/abs/2404.19756 | experiments/backtranslation/papers/70.pdf | The prevailing neural network architecture, the Multi-Layer Perceptron, strictly separates its operations into learnable linear weights on its connections and fixed, simple non-linear activation functions on its nodes. This fundamental design choice appears to be a major bottleneck, leading to poor parameter efficiency... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to produce a research proposal in markdown, with three sections: Introduction, Proposal, Discussion. Must include problem framing, hook, etc. Provide a coherent framework for neural networks that don't separate linear weights and fixed activations. Could propose learnable activation functions paramet... | # Research Proposal
## 1. Introduction
### Research Question
*How can we construct a neural network that is not constrained by the rigid division between learnable linear transformations and fixed, simple non‑linear activation functions?*
### Motivation
The ubiquitous Multi‑Layer Perceptron (MLP) – and by ... |
Ring Attention with Blockwise Transformers for Near-Infinite Context
| Computer Science | Artificial Intelligence (AI) | 2,024 | https://arxiv.org/abs/2310.01889 | experiments/backtranslation/papers/71.pdf | While blockwise computation of attention has made processing longer sequences more memory-efficient within a single accelerator, we are still fundamentally bottlenecked because the full set of layer activations for the entire sequence must be stored, which easily exceeds any single device's memory capacity. Given that ... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to craft a research proposal markdown, about scaling transformer training to extremely long sequences, with memory issue across devices, communication bottlenecks. Need a novel approach.
Idea: hierarchical activation checkpointing combined with activation offloading and asynchronous pipelined recomp... | # Research Proposal
**Title:** *Distributed Invertible Chunked Attention (DICA): A Reversible, Activation‑Compressed Framework for Training Transformers on Ultra‑Long Sequences*
---
## 1. Introduction
### 1.1 Research Question
> **How can we train transformer models on sequences whose activations are far... |
Tulu 3: Pushing Frontiers in Open Language Model Post-Training
| Computer Science | Artificial Intelligence (AI) | 2,024 | https://arxiv.org/abs/2411.15124 | experiments/backtranslation/papers/72.pdf | The capabilities of open-source language models consistently lag behind leading proprietary systems, primarily because the complex post-training recipes and data used by top commercial labs remain undisclosed. Current open-source efforts, which rely on applying simpler tuning pipelines to a patchwork of available publi... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to produce a research proposal in markdown, following the three-part format: Introduction, Proposal, Discussion. Must contain problem framing, hook, precision, derivation, novelty, evidence, testability, implementation, impact, feasibility, falsification, scope, verification, parameter extraction. Mu... | ## 1. Introduction
**Research question**
*How can we build an open‑source language model that matches the reasoning and instruction‑following performance of top proprietary systems, given that the usual post‑training pipelines (static instruction‑tuning on a patchwork of public corpora) appear to have reached a ha... |
Vision Transformers Need Registers
| Computer Science | Artificial Intelligence (AI) | 2,024 | https://arxiv.org/abs/2309.16588 | experiments/backtranslation/papers/73.pdf | We are observing that large-scale vision transformers spontaneously repurpose tokens from visually uninformative background regions for some kind of internal, global computation, causing those tokens to become high-norm outliers and corrupting the feature maps needed for dense prediction tasks. Since the standard archi... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to write a research proposal in markdown with three sections: Introduction, Proposal, Discussion. Need to address the problem: large-scale vision transformers repurpose background tokens for global computation, causing high-norm outliers and corrupting dense prediction. Need solution: prevent sacrifi... | # Research Proposal
## 1. Introduction
### Research Question
*Why do large‑scale Vision Transformers (ViTs) spontaneously repurpose patches that belong to visually uninformative background regions for internal, global‑type calculations?*
The side‑effect is that those background tokens become high‑norm outlier... |
CoLoR-Filter: Conditional Loss Reduction Filtering for Targeted Language Model Pre-training | Computer Science | Artificial Intelligence (AI) | 2,024 | https://arxiv.org/abs/2406.10670 | experiments/backtranslation/papers/74.pdf | Given a massive, general-purpose text corpus and only a small set of examples from a specific target domain, we need to select the most relevant data for pre-training, but current approaches are inadequate; simple filtering based on generic quality heuristics is not specific enough to the target task, while more advanc... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to produce a research proposal in markdown, with three sections: Introduction, Proposal, Discussion. Must include problem framing, hook, precision, derivation, constants/key concepts, novelty, evidence, testability, implementation, impact, feasibility, validation (falsification, scope, verification, ... | # Research Proposal
## 1. Introduction
### Research Question
*Given a massive, general‑purpose text corpus and only a small set of examples from a specific target domain, how can we efficiently select the most relevant training data for domain‑adapted pre‑training without resorting to sequential, computationall... |
How to Train Long-Context Language Models (Effectively)
| Computer Science | Artificial Intelligence (AI) | 2,024 | https://arxiv.org/abs/2410.02660 | experiments/backtranslation/papers/75.pdf | When we attempt to adapt pre-trained language models for longer contexts by continuing their training, we find that simple benchmarks like perplexity or basic fact-retrieval tests provide unreliable signals for real-world task performance, and worse, these adaptation methods often degrade the model's original short-con... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to write a research proposal per given template. Should be markdown, 800-1200 words, include sections Introduction, Proposal, Discussion. Should address problem: adapting pretrained LMs for longer contexts leads to unreliable benchmarks, tradeoff between long-context performance and short-context cap... | # Research Proposal
**Title:** Dual‑Scale Mixed‑Objective Fine‑Tuning (DS‑MOFT) – a principled training regime that enlarges context window while explicitly preserving short‑context competence
---
## 1. Introduction
### 1.1. Research Question
> *When we continue‑training large pre‑trained language models (P... |
SimPO: Simple Preference Optimization with a Reference-Free Reward
| Computer Science | Artificial Intelligence (AI) | 2,024 | https://arxiv.org/abs/2405.14734 | experiments/backtranslation/papers/76.pdf | Direct preference optimization requires a reference model to calculate rewards based on a likelihood ratio, which is not only computationally expensive but is also fundamentally misaligned with the simple log-likelihood used to guide text generation at inference time. Given this practical inefficiency and theoretical d... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to write a research proposal in markdown, with the three-part format: Introduction, Proposal, Discussion. The content: formulating a preference learning objective that directly optimizes generation likelihood without a reference model. Need hook, novelty, etc. Should be concise, detailed, include der... | # Research Proposal
## 1. Introduction
### Research Question
*How can we formulate a preference‑learning objective that directly optimizes the generation likelihood of a language model, without invoking an external reference model to compute rewards?*
### Motivation
Current Direct Preference Optimization (... |
Sheared LLaMA: Accelerating Language Model Pre-training via Structured Pruning
| Computer Science | Artificial Intelligence (AI) | 2,024 | https://arxiv.org/abs/2310.06694 | experiments/backtranslation/papers/77.pdf | Creating smaller, general-purpose language models from larger ones is an appealing, cost-effective alternative to training from scratch, but our current methods for removing model components result in irregular architectures that are computationally inefficient. These shrunken models also exhibit uneven knowledge loss ... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to create a research proposal in markdown with three sections: Introduction, Proposal, Discussion. Must include all required elements: Problem framing, hook, precision, derivation, constants/key concepts definitions, novelty, evidence, testability, implementation, impact, feasibility, falsification, ... | # Research Proposal
## 1. Introduction
### 1.1 Research Question
*How can we produce a high‑quality, smaller‑scale, general‑purpose language model from a large pretrained model while (i) preserving a **regular** (i.e., computationally efficient) architecture and (ii) avoiding uneven loss of knowledge across da... |
SOAP: Improving and Stabilizing Shampoo using Adam
| Computer Science | Artificial Intelligence (AI) | 2,024 | https://arxiv.org/abs/2409.11321 | experiments/backtranslation/papers/78.pdf | Second-order optimizers like Shampoo offer faster convergence than first-order methods but are prohibitively expensive due to the need to constantly recompute their preconditioner matrices. Since simply reducing the update frequency of these matrices leads to significant performance degradation, how can we stabilize th... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to produce a research proposal with specified sections. The question: second-order optimizers like Shampoo expensive due to preconditioner updates; reducing frequency hurts performance. Need to stabilize optimization and effectively use geometric info during long intervals between updates.
We need t... | # 1. Introduction
**Research question**
*Second‑order optimizers such as Shampoo converge faster than first‑order methods, but their preconditioner matrices must be recomputed at every iteration, which is prohibitively expensive. Simply reducing the update frequency of these matrices causes a large drop in perform... |
How Does Critical Batch Size Scale in Pre-training?
| Computer Science | Artificial Intelligence (AI) | 2,024 | https://arxiv.org/abs/2410.21676 | experiments/backtranslation/papers/79.pdf | Existing compute-optimal training strategies require scaling up model parameter count and dataset size in tandem, but this approach makes it impossible to tell if the point of diminishing returns for data parallelism is driven by the larger model or the larger dataset. How can we systematically isolate the individual c... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to produce a research proposal in markdown, with the given structure. Must include problem framing, hook, precision, derivation, constants/key concepts, novelty sentence, evidence, testability, implementation, impact, feasibility, falsification, scope, verification, parameter extraction.
We need to ... | # Research Proposal
**Title:** *Model‑Dataset Disentanglement (MDD) – a systematic framework for isolating the individual influence of model size and data size on the scaling of the efficient batch size*
---
## 1. Introduction
### Research Question
> Existing compute‑optimal training strategies require scali... |
Length-Controlled AlpacaEval: A Simple Way to Debias Automatic Evaluators
| Computer Science | Artificial Intelligence (AI) | 2,024 | https://arxiv.org/abs/2404.04475 | experiments/backtranslation/papers/80.pdf | Automated evaluation using a powerful language model as a judge is a cost-effective way to rank models, but its results are systematically corrupted by a strong preference for longer outputs, a flaw that makes the rankings easy to manipulate. Since we cannot directly modify the behavior of the proprietary judging model... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to write a research proposal according to the template with Introduction, Proposal, Discussion, meeting all required sections: problem framing, hook, precision, derivation, constants, novelty, evidence, testability, implementation, impact, feasibility, validation requirements including falsification,... | ## 1. Introduction
**Research question.**
*How can we mathematically adjust the raw preference scores produced by a proprietary language‑model judge so that the final ranking of candidate systems is not confounded by a simple difference in response verbosity?*
**Motivation.**
Automated evaluation with a power... |
Perfect Zero-Knowledge PCPs for #P
| Computer Science | Theoretical CS | 2,024 | https://arxiv.org/abs/2403.11941 | experiments/backtranslation/papers/123.pdf | The standard sumcheck protocol, when compiled into a non-interactive proof for a counting problem, inherently reveals computationally hard information, making it fundamentally incompatible with perfect zero-knowledge. Given that existing techniques for creating zero-knowledge proofs are limited to statistical privacy a... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to write a research proposal in Markdown, with three parts: Introduction, Proposal, Discussion. Must address the given question: constructing a perfectly private zero-knowledge proof for a hard counting problem, with a non-adaptive verifier, overcoming limitations of sumcheck revealing hard info. Nee... | # Research Proposal
## 1. Introduction
### 1.1 Research Question
*How can we construct a non‑interactive proof for a computationally hard counting problem (e.g., \#SAT) that is **perfectly** zero‑knowledge **and** can be verified by a **non‑adaptive** verifier, despite the inherent leakage of the standard sum‑c... |
Two prover perfect zero knowledge for MIP* | Computer Science | Theoretical CS | 2,024 | https://arxiv.org/abs/2404.00926 | experiments/backtranslation/papers/124.pdf | We know how to construct perfect zero-knowledge proofs for multiprover systems with entanglement, but our only established method relies on distributing the verifier's information across six different provers, a technique that seems fundamentally incompatible with a two-prover setup. Given that the full computational p... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to produce a research proposal in markdown, following the specified template: Introduction, Proposal, Discussion. Need to address the given question: achieve perfect zero-knowledge for two-prover entangled systems without splitting verifier's information across six provers.
We need to present a nove... | # Research Proposal
**Title:** *Quantum‑Secret‑Sharing Zero‑Knowledge for Two‑Prover Entangled Systems*
---
## 1. Introduction
### 1.1 Research Question
Current constructions of **perfect zero‑knowledge (PZK)** for multiprover interactive proofs with entangled provers (MIP*) rely on **splitting the verifier’... |
Quasi-Linear Size PCPs with Small Soundness from HDX
| Computer Science | Theoretical CS | 2,024 | https://arxiv.org/abs/2407.12762 | experiments/backtranslation/papers/125.pdf | Current methods for creating efficient probabilistically checkable proofs are stuck: standard parallel repetition can drive the soundness error to be arbitrarily small but causes a polynomial blow-up in proof size, whereas the size-efficient graph powering technique has a known theoretical barrier that prevents it from... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>The user wants a research proposal in pure Markdown, with three sections: Introduction, Proposal, Discussion. They want a concrete idea for constructing a 2-query PCP (probabilistically checkable proof) with arbitrarily low constant soundness and quasi-linear proof size increase. Need to explain problem, nov... | # Research Proposal
## 1. Introduction
### Research Question
*Can we build a **2‑query probabilistically checkable proof (PCP)** whose soundness error can be reduced to any desired constant **ε > 0**, while the total proof length grows only **quasi‑linearly** (i.e., by a factor *N·polylog N*)?*
### Why the p... |
On the Locality of Hall's Theorem | Computer Science | Theoretical CS | 2,025 | https://arxiv.org/abs/2501.03649 | experiments/backtranslation/papers/126.pdf | For many fundamental distributed graph problems, our best deterministic algorithms run in polylogarithmic time, often relying on general but slow techniques like network decomposition, while the strongest lower bounds remain merely logarithmic. This suggests that while nodes can gather information from a logarithmic-si... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to write proposal in markdown with three sections introduction, proposal, discussion. Should address problem: deterministic polylog algorithms vs lower bounds, need method for globally consistent decisions from local view without extra rounds. Propose new framework: "locally certified decisions via d... | ## 1. Introduction
**Research question.**
In the LOCAL model of distributed computing, many cornerstone graph problems (maximal independent set, (Δ+1)‑coloring, edge‑orientation, etc.) admit *deterministic* algorithms that run in poly‑logarithmic rounds. The standard way to obtain such algorithms is to first buil... |
Asymptotically Optimal Hardness for k-Set Packing | Computer Science | Theoretical CS | 2,025 | https://arxiv.org/abs/2409.17831 | experiments/backtranslation/papers/127.pdf | For fundamental problems like k-Set Packing, the best algorithms achieve an O(k) approximation factor, yet the strongest known hardness-of-approximation results only establish a bound of Ω(k/log k). Since this gap appears to be an artifact of the standard proof technique, where reductions from constraint satisfaction p... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to draft a research proposal with the required sections. The problem is about establishing tighter hardness of approximation for k-Set Packing (k-Set Packing has O(k) approximation algorithms, hardness known is Ω(k / log k). The standard reductions from CSP introduce log factor blowup. Need new reduc... | # Research Proposal
**Title:** A Gap‑Preserving, Log‑Free Reduction Framework for Linear Hardness of *k‑Set Packing*
**Authors:** [Your Name], [Collaborators]
**Date:** September 2025
---
## 1. Introduction
### 1.1 Research Question
*Why does the best known NP‑hardness for *k‑Set Packing* fall short... |
PACE: Poisoning Attacks on Learned Cardinality Estimation
| Computer Science | Databases and Information Management | 2,024 | https://arxiv.org/abs/2409.15990 | experiments/backtranslation/papers/128.pdf | Learned cardinality estimators in database systems are continuously retrained on the historical query workload, but this retraining process itself presents a security flaw. Given that an attacker has no internal access to the model's architecture or parameters and that any malicious queries must be statistically simila... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to write a research proposal in markdown, following template: Introduction, Proposal, Discussion. Must include problem framing, concise hook, precision, derivation, constants/key concepts, novelty, evidence, testability, implementation, impact, feasibility, falsification, scope, verification, paramet... | # Research Proposal
**Title:** *Distribution‑Constrained Influence‑Based Query Poisoning for Learned Cardinality Estimators*
---
## 1. Introduction
### 1.1 Research Question
> *How can an attacker, who lacks internal knowledge of a learned cardinality estimator (LCE) and must issue queries that are statisti... |
Black Hole Singularity Resolution in Unimodular Gravity from Unitarity | Physics | General relativity | 2,024 | https://arxiv.org/abs/2409.03006 | experiments/backtranslation/papers/21.pdf | Quantum models of black hole interiors show that singularity resolution depends on the choice of a time variable, yet for any physically relevant clock that reaches the singularity in a finite duration, the evolution cannot be unitary. Since general relativity offers no preferred time coordinate, how can we formulate a... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to produce a research proposal in markdown, following template with three parts: Introduction, Proposal, Discussion. Must include problem framing, hook, precision, derivation, constants/key concepts, novelty, evidence, testability, implementation, impact, feasibility, validation (falsification, scope... | ## 1. Introduction
### Research question
*Quantum models of black‑hole interiors show that singularity resolution depends on the choice of a time variable, yet for any physically relevant clock that reaches the singularity in a finite duration the evolution cannot be unitary. Since general relativity offers no pre... |
Primordial Black Holes with QCD Color Charge | Physics | General relativity | 2,023 | https://arxiv.org/abs/2310.16877 | experiments/backtranslation/papers/22.pdf | While primordial black holes are theorized to form in the early universe's quark-gluon plasma, our understanding of color-charged black holes is limited to vacuum solutions that do not account for a realistic formation environment. Given that the plasma's inherent Debye screening mechanism should effectively neutralize... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to write a research proposal in markdown with three sections: Introduction, Proposal, Discussion. Must address the question about forming a color-charged black hole in a quark-gluon plasma despite Debye screening. Provide hook, novelty, testability, etc. No equations unless inherently mathematical; c... | # Research Proposal
**Title:** *Color‑Retaining Collapse in a Quark‑Gluon Plasma: A Dynamical‑Screening Framework for Macroscopic Color‑Charged Black Holes*
---
## 1. Introduction
### 1.1. Research Question
> **How can a gravitationally collapsing region of the early‑Universe quark‑gluon plasma (QGP) form a ... |
Gravity from entropy | Physics | General relativity | 2,024 | https://arxiv.org/abs/2408.14391 | experiments/backtranslation/papers/23.pdf | An entropic action based on the relative information between the spacetime metric and a metric induced by a simple scalar field is an appealing concept, yet this approach is fundamentally flawed as it fails to determine the geometry of spacetime in a vacuum and cannot account for massive fields. How can we define a ric... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to write proposal with markdown, sections Introduction, Proposal, Discussion. Must include hook, novelty etc. Must be concise but thorough. Avoid equations unless necessary; could be conceptual. Provide testable predictions. We'll outline an entropic principle that includes both geometry of vacuum an... | # Research Proposal
**Title:** *A Multi‑Field Relative‑Entropy Action for Vacuum Geometry and Massive Matter*
---
## 1. Introduction
### 1.1 Research Question
> *How can we formulate an entropic variational principle that determines the geometry of empty spacetime **and** correctly incorporates massive fiel... |
Cold Dark Matter Based on an Analogy with Superconductivity
| Physics | Solid-state Physics | 2,024 | https://arxiv.org/abs/2408.08356 | experiments/backtranslation/papers/24.pdf | The standard thermal freeze-out model successfully predicts a relic abundance for dark matter, but extensive experimental searches for particles matching this production mechanism have yielded no results, creating a significant tension. How could a population of particles that decoupled from the cosmic plasma very earl... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to produce research proposal in markdown exactly with three parts: Introduction, Proposal, Discussion. Must include required elements: problem framing, hook, precision, derivation, constants, novelty, evidence, testability, implementation, impact, feasibility, falsification, scope, verification, para... | # Research Proposal
## 1. Introduction
**Research question** – *How can a population of particles that decoupled from the primordial plasma extremely early evolve, without any Standard‑Model (SM) interactions, into a state that naturally reproduces the observed cold‑dark‑matter (CDM) density today?*
### Motiva... |
The expected kinematic matter dipole is robust against source evolution
| Physics | Solid-state Physics | 2,024 | https://arxiv.org/abs/2404.07929 | experiments/backtranslation/papers/25.pdf | The standard formula for predicting the kinematic dipole in galaxy counts is being challenged because it doesn't explicitly model how the properties of cosmic sources change with distance, leading some to argue that any calculation must rely on the full, and currently unknown, evolutionary history of the source populat... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to write a research proposal. Must include intro, proposal, discussion sections. Need to present idea that the simple dipole formula using observable slope is mathematically equivalent to the full evolution-dependent calculation. Provide reasoning, testable predictions. Include novelty, etc.
Will in... | ## 1. Introduction
**Research question** – *Is the traditional kinematic‑dipole estimator for galaxy counts, which depends only on the observed slope of the differential source‑count distribution at the survey flux limit, fundamentally equivalent to a fully evolution‑aware calculation that integrates the unknown cos... |
Probing general relativistic spin-orbit coupling with gravitational waves from hierarchical triple systems | Physics | Solid-state Physics | 2,024 | https://arxiv.org/abs/2307.01903 | experiments/backtranslation/papers/26.pdf | The standard geometrical optics framework for modeling lensed gravitational waves treats them as rays following null geodesics, an approach that completely neglects the wave's internal polarization and thus cannot account for any interaction between polarization and spacetime curvature. How would a wave packet's trajec... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to write a research proposal following the specified template. Include problem framing, hook, precision, derivation, novelty, evidence, testability, implementation, impact, feasibility, falsification, scope, verification, parameter extraction. Should be about ~800-1200 words, markdown. No equations u... | # Research Proposal
**Title:** *Polarization‑Dependent Propagation of Gravitational‑Wave Packets in Curved Spacetime – A Spin‑Hall Effect for Gravitational Waves*
---
## 1. Introduction
### 1.1 Research Question
> *How does the internal polarization of a gravitational‑wave (GW) packet influence its trajector... |
Global Networks, Monetary Policy and Trade | Economics | Macroreconomics | 2,025 | https://www.suomenpankki.fi/contentassets/9b4df918cdd640b7a92a71588ff209b5/4-sebnem-kalemli-ozcan.pdf | experiments/backtranslation/papers/177.pdf | The body of research connecting macroeconomics to climate change has expanded into numerous specialized sub-fields, from calculating the social cost of carbon to modeling green innovation and adaptation strategies, yet these topics are often studied in isolation using distinct frameworks. Without a clear organizing pri... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to produce a research proposal in markdown with three-part format: Introduction, Proposal, Discussion. Need to satisfy all essential requirements: problem framing, hook, precision, derivation, key concepts, novelty, evidence, testability, implementation, impact, feasibility, falsification, scope, ver... | # **Research Proposal**
## 1. Introduction
### 1.1. Research Question
*How can we structure the rapidly expanding, fragmented macro‑economic–climate literature into a unified conceptual map that makes the system‑wide trade‑offs and inter‑dependencies explicit and testable?*
### 1.2. Motivation
Over the past... |
Macroeconomics and Climate Change | Economics | Macroreconomics | 2,025 | https://www.nationalacademies.org/documents/embed/link/LF2255DA3DD1C41C0A42D3BEF0989ACAECE3053A6A9B/file/D4D69AD2B6747593022189BE9D1F36B8A37D22C82040?noSaveAs=1 | experiments/backtranslation/papers/178.pdf | Economic theory suggests that as wage gaps widen among similar workers, higher earners may prefer individual negotiation over collective bargaining, thereby undermining union solidarity. However, empirical studies have only established a correlation between rising inequality and declining unionization, making it imposs... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to write a research proposal in markdown, following the template. Need to address causal determination of wage dispersion effect on union support. Provide a coherent framework, maybe using a natural experiment, a causal inference design like regression discontinuity, instrumental variables, or a fiel... | # Research Proposal
## 1. Introduction
### Research Question
*Does greater wage dispersion among otherwise similar workers *directly* erode support for the labour movement, or is the observed correlation merely a by‑product of weak unions failing to curb inequality?*
### Motivation
Economic theory predicts... |
How Does Wage Inequality Affect the Labor Movement? | Economics | Microeconomics | 2,025 | https://zcullen.github.io/assets/docs/Labor_Movement_WP.pdf | experiments/backtranslation/papers/179.pdf | Sellers increasingly operate in a dual-channel environment, selling through their own websites while also using data-rich digital platforms that can perfectly match products to individual consumers. Since existing economic models of competition do not fully address the fact that a consumer can use the platform's person... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>The user wants a research proposal in markdown, with three sections: Introduction, Proposal, Discussion. Must include problem framing, hook, precision, derivation, constants, novelty, evidence, testability, implementation, impact, feasibility, falsification, scope, verification, parameter extraction, etc.
T... | # **Research Proposal**
**Title:** *Channel‑Aware Pricing‑Quality (CAPQ) Framework for Dual‑Channel Sellers Facing Showrooming*
---
## 1. Introduction
### Research Question
How do sellers who operate both a proprietary web store and a data‑rich digital platform (e.g., Amazon, Google Shopping) optimally choos... |
Data, Competition, and Digital Platforms
| Economics | Microeconomics | 2,024 | https://arxiv.org/abs/2304.07653 | experiments/backtranslation/papers/180.pdf | Standard audit studies measure direct bias by comparing outcomes for individuals who have identical qualifications, but this approach explicitly ignores how discrimination in an earlier stage—like a previous job or in the justice system—can create the very differences in qualifications that the experiment holds constan... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to produce a research proposal in markdown with three sections: Introduction, Proposal, Discussion. Must meet all essential requirements: problem framing, concise hook, precision, derivation, constants/key concepts, novelty, evidence, testability, implementation, impact, feasibility, falsification, s... | # Research Proposal
**Title:** *Cumulative Discrimination Audits – Measuring the Total Impact of Past and Present Bias through a Two‑Stage Causal Framework*
---
## 1. Introduction
### 1.1 Research Question
Standard audit (field‑experiment) studies estimate **direct** discrimination by holding applicants’ ob... |
Systemic Discrimination: Theory and Measurement | Economics | Microeconomics | 2,024 | https://haslam.utk.edu/wp-content/uploads/2023/02/Systemic_Discrimination_Nov15_2022.pdf | experiments/backtranslation/papers/181.pdf | Decades of economic research show that technological advances primarily augment the productivity of high-skilled workers by automating routine tasks, thereby widening the performance gap between experts and novices. Since new systems can learn from the actions of all workers and potentially codify and distribute the su... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to produce a research proposal in markdown, following the three-part format: Introduction, Proposal, Discussion. Must include all required sections: problem framing, hook, precision, derivation, constants/key concepts, novelty, evidence, testability, implementation, impact, feasibility, falsification... | # Research Proposal
## 1. Introduction
### Research Question
*How does providing workers at all skill levels with an AI‑driven tool that learns from the actions of every employee and codifies the subtle, successful behaviors of top performers affect productivity in a real‑world business environment?*
### Mot... |
Generative AI at Work | Economics | Microeconomics | 2,024 | https://arxiv.org/abs/2304.11771 | experiments/backtranslation/papers/182.pdf | Standard models of shareholder voting analyze corporate decisions by assuming the group of voters is fixed, but this fails to account for the fact that liquid stock markets allow investors to trade into or out of a company based on their alignment with a pending proposal. Since this pre-vote trading actively shapes the... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to output markdown with three sections Introduction, Proposal, Discussion. Need to follow requirements: hook, novelty, testability, etc. The idea: dynamic shareholder voting where pre-vote trading changes electorate. We'll propose a framework: "Endogenous Electorate Model" combining voting games with... | # Research Proposal
## 1. Introduction
### Research Question
*How can we properly analyse the outcomes of shareholder democracy when investors are free to trade into or out of a company in anticipation of a pending proposal, thereby turning the electorate itself into a strategic, endogenous variable?*
### Wh... |
Trading and Shareholder Democracy | Economics | Finance | 2,024 | https://www.ecgi.global/sites/default/files/working_papers/documents/tradingfinal.pdf | experiments/backtranslation/papers/183.pdf | Economic theory has long held that a bank's core function of transforming illiquid assets into liquid deposits makes it inherently vulnerable to panic-based runs, yet large-scale empirical studies have been unable to confirm this because they cannot disentangle depositor withdrawals driven by worsening fundamentals fro... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to write a research proposal in markdown, following the specified template: Introduction, Proposal, Discussion. Must include problem framing, hook, precision, derivation, key concepts, novelty, evidence, testability, implementation, impact, feasibility, falsification, scope, verification, parameter e... | # Research Proposal
## 1. Introduction
### Research Question
*How can we use broad, high‑frequency banking data to isolate and measure the *panic‑driven* fragility that theory predicts for banks, while rigorously separating it from withdrawals that are purely a reaction to worsening fundamentals?*
### Motiv... |
Liquidity Transformation and Fragility in the U.S. Banking Secto | Economics | Finance | 2,024 | https://www.nber.org/system/files/working_papers/w27815/w27815.pdf | experiments/backtranslation/papers/184.pdf | Traded dividend strips allow for direct measurement of the equity term structure, but the available data is severely limited, covering only a short time period for the aggregate market and a few maturities. Given that alternative methods based on stock portfolios require imposing strong and potentially unrealistic theo... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to produce a research proposal in markdown with three sections: Introduction, Proposal, Discussion. Must include all required elements: problem framing, concise hook, precision, derivation, constants/key concepts, novelty, evidence, testability, implementation, impact, feasibility, falsification, sco... | # Research Proposal
## 1. Introduction
### 1.1 Research Question
*How can we generate reliable, long‑run dividend term‑structure estimates for a wide variety of equity portfolios when traded dividend strips are scarce and alternative portfolio‑based approaches rely on strong, possibly unrealistic, theoretical ... |
Equity Term Structures without Dividend Strips Data | Economics | Finance | 2,024 | https://www.nber.org/system/files/working_papers/w31119/w31119.pdf | experiments/backtranslation/papers/185.pdf | To accurately measure a portfolio’s 'abnormal price'—the gap between its market price and its fundamental value according to an asset pricing model—we cannot simply use short-horizon abnormal returns, as these do not properly account for price levels. However, the alternative of directly discounting long-run future cas... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to produce a research proposal per template, addressing the problem of measuring abnormal price (mispricing) that captures price level differences, avoiding short-horizon abnormal returns and long-run cash flow discounting. Need a robust estimator. Provide hook, novelty, testability, implementation, ... | ## 1. Introduction
### Research Question
*How can we construct a robust estimator of a portfolio’s **abnormal price** – the gap between its market price and its model‑implied fundamental value – that (i) respects price‑level information, (ii) avoids the bias‑prone long‑run cash–flow discounting exercise, and (iii)... |
Putting the Price in Asset Pricing | Economics | Finance | 2,024 | https://personal.lse.ac.uk/polk/research/PuttingThePriceInAP.pdf | experiments/backtranslation/papers/186.pdf | Financial regulators typically design stress tests using a single, plausible adverse scenario to determine if banks have enough capital, but this method offers no formal guidance on how to choose or combine different hypothetical shocks to best uncover hidden vulnerabilities. Since the goal is to gather information to ... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to produce research proposal in markdown with three sections: Introduction, Proposal, Discussion. Must address systematic construction of stress-test scenarios to maximize learning about banks' hidden risk exposures. Need hook, novelty, testability, etc. No equations unless inherently mathematical; b... | ## 1. Introduction
**Research question**
*How can regulators construct stress‑test scenarios – possibly several, possibly combined – so that the *information* they obtain about each bank’s hidden, unobservable risk exposures is maximised?*
**Why the problem matters**
Regulators currently run a **single, plaus... |
Designing Stress Scenarios | Economics | Finance | 2,024 | https://pages.stern.nyu.edu/~tphilipp/papers/Parlatore_Philippon_Stress_Tests.pdf | experiments/backtranslation/papers/187.pdf | The prevailing strategy for socially responsible investing is to allocate capital to firms with high social value, with the goal of lowering their cost of capital and encouraging their positive work; however, this strategy doesn't account for the fact that many of these firms are already attractive to purely commercial... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to produce a research proposal in markdown with three parts: Introduction, Proposal, Discussion. Follow requirements: Problem framing, hook, precision, derivation, key concepts, novelty, evidence, testability, implementation, impact, feasibility, falsification, scope, verification, parameter extracti... | # Research Proposal
## 1. Introduction
### Research Question
*When does a socially‑oriented investment create **additional** social good, and when does it merely replace commercial capital that would have funded the same project anyway?*
### Motivation
Socially responsible investing (SRI) is usually justif... |
The Allocation of Socially Responsible Capital | Economics | Finance | 2,024 | https://www.hbs.edu/ris/Publication%20Files/AllocationSociallyResponsibleCapital_c5f3104f-8871-4c7d-94fa-68aec509807f.pdf | experiments/backtranslation/papers/188.pdf | Arbitrage pricing theory assumes that arbitrageurs have perfect knowledge of asset mispricing, but in high-dimensional markets with many assets and finite data, any true mispricing signals are often weak and rare, making them fundamentally difficult to distinguish from statistical noise. Given that any feasible strateg... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to write a research proposal in markdown, with three sections: Introduction, Proposal, Discussion, fulfilling all the specifications.
The question: Maximum achievable risk-adjusted return for an arbitrageur constrained by statistical learning problem in high-dimensional markets, where mispricing sig... | # Research Proposal
## 1. Introduction
**Research question** – *What is the maximum achievable risk‑adjusted return (e.g., Sharpe ratio) for an arbitrageur who must rely on imperfect, data‑driven estimates of asset mispricing in a high‑dimensional market where true mispricing signals are weak, rare, and hidden in... |
Multipartite Hardy paradox unlocks device-independent key sharing | Computer Science | Quantum Computing | 2,025 | https://arxiv.org/abs/2505.10170 | experiments/backtranslation/papers/105.pdf | Device-independent conference key agreement protocols are fundamentally limited by their reliance on creating and maintaining genuine multipartite entangled states, which are experimentally fragile and resource-intensive, and the secret key is always extracted from the resulting measurement outcomes. How can we devise ... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to write a research proposal in markdown, with three sections: Introduction, Proposal, Discussion. Must include all items: problem framing, hook, precision, derivation, constants/key concepts, novelty, evidence, testability, implementation, impact, feasibility, falsification, scope, verification, par... | # Device‑Independent Conference Key Agreement without Multipartite Entanglement
## 1. Introduction
### Research Question
Current device‑independent (DI) conference key agreement (CKA) protocols rely on the creation and preservation of genuine multipartite entangled states. These states are experimentally fragi... |
Barren Plateaus in Variational Quantum Computing
| Computer Science | Quantum Computing | 2,024 | https://arxiv.org/abs/2405.00781 | experiments/backtranslation/papers/106.pdf | Variational quantum algorithms consistently fail to train as the number of qubits increases because their loss function gradients appear to shrink exponentially, rendering classical optimization practically impossible. Since this severe training obstruction appears for many different choices of circuits and measurement... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to produce a research proposal following template. Need to explain underlying principle why variational quantum algorithms (VQAs) suffer from barren plateaus: gradients shrink exponentially. The principle is concentration of measure in high-dimensional Hilbert space, leading to gradient variance vani... | # **Research Proposal**
**Title:** *Measure‑Concentration as the Unifying Principle Behind Barren Plateaus in Variational Quantum Algorithms*
---
## 1. Introduction
### 1.1 Motivation
Variational quantum algorithms (VQAs) – such as the variational quantum eigensolver (VQE) and quantum approximate optimizat... |
Blocklet concatenation: low-overhead fault-tolerant protocols for fusion-based quantum computation | Computer Science | Quantum Computing | 2,025 | https://arxiv.org/abs/2506.13619 | experiments/backtranslation/papers/107.pdf | Prevailing methods for generating fault-tolerant protocols in fusion-based quantum computing, such as foliating stabilizer circuits or using geometric templates, consistently lead to surface-code-like architectures with a resource footprint that grows quadratically with the required error suppression. Given this signif... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to write a research proposal in markdown, following three-part format: Introduction, Proposal, Discussion. Must include all required sections: problem framing, concise hook, precision, derivation, constants/key concepts, novelty, evidence, testability, implementation, impact, feasibility, falsificati... | # Research Proposal
**Title:** *Error‑Absorbing Fusion Tiles and Fractal‑Dimension Fusion Networks – A Sub‑Quadratic Path to Fault‑Tolerant Fusion‑Based Quantum Computing*
---
## 1. Introduction
### Research Question
Current fault‑tolerant protocols for fusion‑based quantum computing (FBQC) rely on **foliate... |
Fault-tolerant quantum simulation via symplectic transvections | Computer Science | Quantum Computing | 2,025 | https://arxiv.org/abs/2504.11444 | experiments/backtranslation/papers/108.pdf | The standard method for fault-tolerant quantum simulation requires breaking down structured algorithms like Trotter circuits into a sequence of individual logical gates, a process where each step incurs significant resource overhead and fails to leverage the algorithm's inherent global structure. How can we devise a fa... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to write research proposal in markdown, following sections Introduction, Proposal, Discussion. Must include problem framing, hook, precision, derivation, constants, novelty, evidence, testability, implementation, impact, feasibility, falsification, scope, verification, parameter extraction. Should be... | # Research Proposal
## 1. Introduction
### Research question
*How can we construct a fault‑tolerant implementation of structured simulation circuits (e.g., Trotter layers) that treats the whole circuit as a single logical object, thereby avoiding the large gate‑by‑gate overhead of current methods?*
### Motiv... |
Error correction of transversal CNOT gates for scalable surface code computation
| Computer Science | Quantum Computing | 2,024 | https://arxiv.org/abs/2408.01393 | experiments/backtranslation/papers/109.pdf | Applying a transversal CNOT gate between two surface codes creates correlated Pauli errors across the code blocks, meaning standard minimum-weight perfect matching decoders are no longer effective as they assume errors are independent within a single code. Since current alternatives are either suboptimal, like combinin... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>The user wants a research proposal in Markdown following a specific template with three parts: Introduction, Proposal, Discussion. Must include all required sections like problem framing, hook, precision, derivation, novelty, evidence, testability, implementation, impact, feasibility, falsification, scope, v... | ## 1. Introduction
**Research question** – *How can we exploit the known cross‑code Pauli‐error correlations introduced by a transversal CNOT between two surface‑code blocks, while still using a decoder that has the speed and scalability of conventional minimum‑weight perfect‑matching (MWPM) decoders?*
### Motiva... |
Scaling Advantage in Approximate Optimization with Quantum Annealing
| Computer Science | Quantum Computing | 2,024 | https://arxiv.org/abs/2401.07184 | experiments/backtranslation/papers/110.pdf | Quantum annealers have yet to show a clear scaling advantage over the best classical optimization heuristics, largely because inherent analog noise and control errors make finding the exact ground state for large, complex problems impractical. Given that the demand for a perfect solution seems to be the primary barrier... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to write a research proposal in markdown, with three sections Introduction, Proposal, Discussion. The idea: reformulate goal of computation to test for advantage on imperfect quantum annealers. We need hook, novelty, testability, etc. Should be concise but thorough, 800-1200 words. No forced equation... | # **Research Proposal**
## 1. Introduction
### 1.1 Motivation & Gap
Quantum annealers (QAs) — most notably the D‑Wave family — promise a physical route to solving large combinatorial optimisation problems by slowly driving a quantum system toward its ground state. Decades of benchmarking, however, have shown **... |
Rethinking Being and Time as a Resource for Feminist Philosophy | Humanities | Philosophy | 2,024 | https://philpapers.org/archive/KNORBA-2.pdf | experiments/backtranslation/papers/153.pdf | Feminist theory requires a framework that can account for the concrete, lived experience of gender and the oppressive nature of social roles, yet influential phenomenological accounts of human existence seem to fail this task by positing a universal, abstract subject that erases difference. Given that this philosophica... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to write a research proposal in markdown with intro, proposal, discussion sections. Should address the problem: phenomenology's universal subject vs feminist need for situated subject. Propose a framework integrating phenomenology with feminist standpoint theory, maybe "Situated Phenomenology" using ... | # **Research Proposal:** *Situated Phenomenology – A Feminist‑Grounded Framework for Analyzing the Socially and Politically Situated Subject*
---
## 1. Introduction
### Research Question
*How can we develop a robust analysis of the socially and politically situated subject that retains phenomenology’s focus on... |
The Screens of Virtual Production: What Is Real? | Humanities | Movies | 2,025 | https://library.oapen.org/bitstream/handle/20.500.12657/100177/9781003463139_10.4324_9781003463139-15.pdf;jsessionid=C31E665E644487B0A73965B7582B9D8D?sequence=1 | experiments/backtranslation/papers/154.pdf | New character creation tools use photogrammetric scans of real people to offer unprecedented diversity, seemingly solving issues of representation by treating human features as a vast, objective database. Since this data-driven approach reduces identity to a set of mathematically blendable traits and sidesteps the soci... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to write a research proposal addressing the given question about photogrammetric scans leading to depoliticized definition of race. Must follow 3-part format: Introduction, Proposal, Discussion. Must include all required sections: Problem framing, hook, precision, derivation etc. Ensure not to use ma... | ## 1. Introduction
### 1.1 Research question & motivating problem
*How does the technical architecture of photogrammetric character‑creation pipelines construct a new, depoliticised definition of race?*
Current game‑engine and virtual‑avatar pipelines increasingly rely on large‑scale photogrammetric scans of re... |
Institutions, Technology and Prosperity | Humanities | Sociology | 2,025 | https://economics.mit.edu/sites/default/files/2025-01/Institutions%2C%20Technology%20and%20Prosperity.pdf | experiments/backtranslation/papers/155.pdf | While we consistently observe that former colonies with better institutions are wealthier, we cannot prove this is a causal relationship, as wealth itself might enable better institutions or some other historical factor could be driving both. Given that standard statistical methods cannot untangle this, how can we iden... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to write a research proposal. Must be markdown. Follow three-part template: Introduction, Proposal, Discussion. Need to include hook, novelty, etc. Should not force math; we can use conceptual approach. The question: need a historical factor that influenced institutions but is otherwise unrelated to ... | # Research Proposal
## 1. Introduction
### Research Question
*Why do former colonies with “better” institutions tend to be richer, and can we isolate a truly exogenous historical factor that shaped those institutions without being itself a driver of later economic performance?*
### Motivation
The empirical... |
AI and Social Media: A Political Economy Perspective | Humanities | Sociology | 2,025 | https://economics.mit.edu/sites/default/files/2025-05/AI%20and%20Social%20Media%20-%20A%20Political%20Economy%20Perspective.pdf | experiments/backtranslation/papers/156.pdf | We observe that political polarization is rising concurrently with the proliferation of online platforms whose business models depend on maximizing user engagement through algorithmic content curation and targeted advertising. Yet, existing political economy models do not provide a formal mechanism to explain how a pla... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to produce a research proposal in markdown, with three sections: Introduction, Proposal, Discussion. Must include problem framing, concise hook, precision, derivation, novelty, evidence, testability, implementation, impact, feasibility, falsification, scope, verification, parameter extraction.
Need ... | ## 1. Introduction
### Research Question
*Why do profit‑maximising decisions of online platforms—particularly the choice of monetisation strategy and the design of user‑interaction mechanisms—systematically generate greater ideological division among citizens and political parties?*
### Motivation
Political p... |
The Urgent Need to Tax Digital Advertising | Humanities | Sociology | 2,025 | https://shapingwork.mit.edu/wp-content/uploads/2024/04/Digital-Ad-Tax_Policy-Brief.pdf | experiments/backtranslation/papers/157.pdf | The dominant business model for major internet platforms, which relies on maximizing user engagement for advertising revenue, appears to be the direct cause of significant social harms like the spread of misinformation and political polarization. Since standard policy tools like antitrust enforcement would likely fail ... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to produce a research proposal markdown with three sections: Introduction, Proposal, Discussion. Must include problem framing, hook, precision, derivation, constants/key concepts, novelty, evidence, testability, implementation, impact, feasibility, falsification, scope, verification, parameter extrac... | ## 1. Introduction
**Research question**
*How can policy reshape the economic incentives of dominant internet platforms so that they are no longer driven by “attention‑at‑all‑costs” and the resulting social harms (misinformation, polarization), when traditional antitrust tools are insufficient?*
**Motivation** ... |
Misperceptions and Demand for Democracy under Authoritarianism | Humanities | Sociology | 2,025 | https://economics.mit.edu/sites/default/files/2024-09/Misperceptions%20and%20Demand%20for%20Democracy%20under%20Authoritarianism_0.pdf | experiments/backtranslation/papers/158.pdf | Support for long-standing authoritarian governments often persists even when they perform poorly on critical issues like disaster relief and corruption, and it is commonly assumed that these supporters are either ideologically committed or completely persuaded by state propaganda. Given that typical information campaig... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to produce research proposal in markdown with three sections: Introduction, Proposal, Discussion. Must include problem framing, hook, precision, derivation, constants/key concepts, novelty, evidence, testability, implementation, impact, feasibility, falsification, scope, verification, parameter extra... | ## 1. Introduction
### Research question
Why do large segments of the electorate continue to back long‑standing authoritarian regimes even when those regimes visibly fail on issues that matter to citizens (e.g., disaster relief, corruption)? Are these supporters *ideologically locked* to the regime, or do they si... |
Universality classes for the coalescent structure of heavy-tailed Galton-Watson trees
| Mathematics | Probability | 2,023 | https://arxiv.org/abs/2302.02960 | experiments/backtranslation/papers/39.pdf | The ancestral structure for critical branching populations is well-understood when the offspring distribution has finite variance, consistently leading to genealogical trees built from a sequence of pairwise mergers. This existing framework, however, cannot handle populations with heavy-tailed offspring distributions w... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to produce a research proposal in markdown, following template: Introduction, Proposal, Discussion.
Need to answer question: universal form of the coalescent process for heavy-tailed offspring distributions (critical branching). Need to propose a framework, likely using Λ-coalescents or Beta-coalesc... | # Research Proposal
## 1. Introduction
### 1.1 Research Question
*What is the universal form of the genealogical (coalescent) process that describes critical branching populations whose offspring distribution has a heavy tail (infinite variance)?*
In the classical setting, when the offspring variance is fin... |
On the spectral edge of non-Hermitian random matrices
| Mathematics | Probability | 2,024 | https://arxiv.org/abs/2404.17512 | experiments/backtranslation/papers/40.pdf | We have established spectral universality at the edge for specific non-Hermitian random matrices, such as those with Gaussian entries, by relying on their analytical tractability which provides explicit formulas for eigenvalue correlations. This approach completely breaks down for the more general case of a matrix with... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to produce a research proposal in markdown, covering intro, proposal, discussion. The problem: proving universality of edge eigenvalue statistics for non-Hermitian random matrices with independent entries plus deterministic perturbation. Need to propose a method: maybe use local law, Girko hermitizat... | ## 1. Introduction
**Research question.**
For an \(N\times N\) non‑Hermitian matrix
\[
X = A + D ,
\]
where the entries of \(A\) are independent (not necessarily Gaussian) with zero mean, unit variance and uniformly bounded higher moments, and \(D\) is a deterministic matrix of comparable norm, we would like t... |
Sharp Thresholds for Factors in Random Graphs | Mathematics | Probability | 2,024 | https://arxiv.org/abs/2411.14138 | experiments/backtranslation/papers/41.pdf | The powerful coupling method that relates random graphs to random hypergraphs has successfully determined sharp F-factor thresholds for complete graphs, but this approach fails for general strictly 1-balanced graphs because the probability of certain overlapping structures—especially sparse cycles where two copies of t... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We must produce a research proposal in markdown with sections: Introduction, Proposal, Discussion. Must address the problem: coupling random graphs to random hypergraphs works for complete graphs but not for general strictly 1-balanced graphs because overlapping sparse cycles have mismatched probabilities. N... | ## 1. Introduction
**Research question**
*Why does the powerful coupling that turns an \(F\)-factor problem in the random graph \(G(n,p)\) into a perfect‑matching problem in a random \(|F|\)-uniform hypergraph break down for general strictly 1‑balanced graphs, and how can we restore a correspondence despite the mi... |
Concentration and Moment Inequalities for Heavy-Tailed Random Matrices | Mathematics | Probability | 2,024 | https://arxiv.org/abs/2407.12948 | experiments/backtranslation/papers/42.pdf | Existing concentration inequalities for sums of random matrices are powerful but rely on strong assumptions, like bounded norms or sub-exponential tails, which are violated by heavy-tailed distributions where only a few moments of the matrix norm may be finite. Given that the established theories for heavy-tailed scala... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to write a research proposal in markdown using the template. Must include hook, novelty, etc. It can be mathematical. The problem is about deviation bounds for spectral norm of sum of independent random matrices under minimal moment conditions (heavy-tailed). So propose a framework: use truncation + ... | # Research Proposal
## 1. Introduction
### 1.1 Research Question
*How can we obtain non‑asymptotic deviation bounds for the spectral norm of a sum of independent random matrices when the only available information is a finite \(p\)‑th moment of the operator norm (for some modest \(p>2\))?*
Current matrix con... |
New Lower Bounds for the (Near) Critical Ising and φ^4 Models’ Two-Point Functions | Mathematics | Probability | 2,024 | https://arxiv.org/abs/2404.05700 | experiments/backtranslation/papers/43.pdf | For critical ferromagnetic spin models in dimensions four and higher, our theoretical understanding is inconsistent with our rigorous results: the two-point correlation function is expected to decay according to mean-field theory, matching the known infrared upper bound, but the best proven general lower bound implies ... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to draft a research proposal addressing the problem: need stronger non-perturbative lower bound on two-point correlation decay in critical ferromagnetic spin models in d>=4. Provide a coherent framework, testable predictions, etc. Should be ~800-1200 words. Use markdown sections Introduction, Proposa... | # Research Proposal
## 1. Introduction
### 1.1. The problem
For ferromagnetic spin models (Ising, \(\varphi^{4}\), Potts, etc.) at their critical temperature \(T_{c}\) the two‑point correlation function
\[
G(x)=\langle \sigma_{0}\,\sigma_{x}\rangle_{T_{c}}
\]
is believed to decay like the *mean‑field* powe... |
Unramified geometric Langlands conjecture | Mathematics | Algebra | 2,025 | https://arxiv.org/abs/2405.03599 | experiments/backtranslation/papers/44.pdf | The established framework of the spectral action and Whittaker normalization allows for the construction of a "coarse" version of the geometric Langlands functor, but this mapping to the category of quasi-coherent sheaves does not inherently provide the structural information needed for the full conjecture. The formal ... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to produce a research proposal addressing the problem: constructing cohomological bounds for a coarse functor from spectral action/Whittaker normalization producing a coarse geometric Langlands functor to QCoh, and need to lift to IndCoh with bounded cohomology on compact objects. The proposal must o... | ## 1. Introduction
**Research question.**
The “spectral action + Whittaker normalization’’ construction furnishes a **coarse geometric‑Langlands functor**
\[
F_{\text{coarse}}:\;\operatorname{QCoh}(\operatorname{LocSys}_{\check G})\longrightarrow \operatorname{QCoh}(\operatorname{Bun}_G)
\]
which is compatible... |
Visual Autoregressive Modeling: Scalable Image Generation via Next-Scale Prediction | Computer Science | Artificial Intelligence (AI) | 2,024 | https://arxiv.org/abs/2404.02905 | experiments/backtranslation/papers/63.pdf | Autoregressive models excel in language by predicting the next token in a 1D sequence, yet applying this same paradigm to vision by flattening a 2D grid of image tokens into a raster-scan order performs poorly and is computationally inefficient. Since this flattening process inherently breaks the essential spatial stru... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to produce research proposal in markdown, following three-part format: Introduction, Proposal, Discussion. Must include hook, novelty, testability, etc. Must not use LaTeX unless inherent. Provide clear concepts. We propose a new autoregressive ordering that respects 2D grid: perhaps using a 2D causa... | # Research Proposal
## 1. Introduction
### 1.1 Research Question
> Autoregressive models succeed in language because the natural data are 1‑D sequences. When the same paradigm is forced onto images by flattening the 2‑D pixel grid into a raster‑scan sequence, performance drops and training costs explode. **How ... |
Rho-1: Not All Tokens Are What You Need
| Computer Science | Artificial Intelligence (AI) | 2,024 | https://arxiv.org/abs/2404.07965 | experiments/backtranslation/papers/64.pdf | The standard approach to pre-training language models applies a uniform next-token prediction loss to every token in the training corpus, which is computationally wasteful since many tokens are either trivial and already learned or persistently noisy and hard to predict. Given that this brute-force, undifferentiated me... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to write a research proposal in markdown with three sections: Introduction, Proposal, Discussion. Must include hook, precision, novelty, etc. Must be about adapting language model pretraining to focus on most informative tokens, avoiding waste. We can propose an "adaptive token weighting" framework u... | # Research Proposal
**Title:** *Adaptive Informativeness‑Weighted Pre‑training for Large‑Scale Language Models*
---
## 1. Introduction
### 1.1 Research Question
> **How can we modify the next‑token prediction objective so that pre‑training concentrates its computation on the most informative parts of the da... |
Mamba: Linear-Time Sequence Modeling with Selective State Spaces
| Computer Science | Artificial Intelligence (AI) | 2,024 | https://arxiv.org/abs/2312.00752 | experiments/backtranslation/papers/65.pdf | Efficient sequence models like structured state space models offer a linear-time alternative to quadratic-scaling Transformers, but their performance on complex data like language is poor because their underlying linear time-invariant structure prevents them from selectively processing information based on content. How... | <|start|>system<|message|>You are ChatGPT, a large language model trained by OpenAI.
Knowledge cutoff: 2024-06
Current date: 2025-09-16
Reasoning: medium
# Valid channels: analysis, commentary, final. Channel must be included for every message.<|end|><|start|>developer<|message|># Instructions
You are a knowledgeabl... | <|message|>We need to produce a research proposal in markdown with three parts: Introduction, Proposal, Discussion. Must address building a model that retains recurrent efficiency while overcoming fixed dynamics limitation. Provide hook, novelty, testability, etc. Should be about 800-1200 words.
We need to avoid equat... | # Efficient, Content‑Adaptive Recurrent Models for Long‑Sequence Language Understanding
---
## 1. Introduction
### Research question
*Efficient sequence models such as Structured State‑Space (S4) models run in linear time, but their **time‑invariant** dynamics cannot select or amplify information based on the... |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.