id stringlengths 33 34 | doc_url stringlengths 33 34 | title stringlengths 10 211 | publication_date timestamp[s] | update_date timestamp[s] | authors stringlengths 5 20.1k | category_primary stringclasses 78
values | category_all stringlengths 5 138 | abstract stringlengths 123 3.09k | embeddings listlengths 768 768 |
|---|---|---|---|---|---|---|---|---|---|
http://arxiv.org/abs/2406.12030v1 | http://arxiv.org/pdf/2406.12030v1 | SPA-VL: A Comprehensive Safety Preference Alignment Dataset for Vision Language Model | 2024-06-17T18:57:37 | 2024-06-17T18:57:37 | Yongting Zhang, Lu Chen, Guodong Zheng, Yifeng Gao, Rui Zheng, Jinlan Fu, Zhenfei Yin, Senjie Jin, Yu Qiao, Xuanjing Huang, Feng Zhao, Tao Gui, Jing Shao | cs.CV | cs.CV, cs.AI, cs.CL | The emergence of Vision Language Models (VLMs) has brought unprecedented
advances in understanding multimodal information. The combination of textual
and visual semantics in VLMs is highly complex and diverse, making the safety
alignment of these models challenging. Furthermore, due to the limited study on
the safety a... | [
-0.3298496902,
-0.3434074819,
0.1790491194,
0.094726041,
-0.0961677283,
-0.1239910126,
0.1400907338,
-0.5728731155,
-0.2641246617,
0.8914022446,
-0.4294802248,
-0.6835506558,
-0.5682740211,
-0.1936921626,
-0.3935483992,
1.1609034538,
0.1768195778,
0.185285598,
0.1403510123,
-0.... |
http://arxiv.org/abs/2406.12023v1 | http://arxiv.org/pdf/2406.12023v1 | LiLiuM: eBay's Large Language Models for e-commerce | 2024-06-17T18:45:41 | 2024-06-17T18:45:41 | Christian Herold, Michael Kozielski, Leonid Ekimov, Pavel Petrushkov, Pierre-Yves Vandenbussche, Shahram Khadivi | cs.CL | cs.CL, cs.LG | We introduce the LiLiuM series of large language models (LLMs): 1B, 7B, and
13B parameter models developed 100% in-house to fit eBay's specific needs in
the e-commerce domain. This gives eBay full control over all aspects of the
models including license, data, vocabulary, and architecture. We expect these
models to be ... | [
-0.204658553,
-0.7289928198,
0.2584224343,
0.1643953174,
-0.4157935679,
0.0487751216,
-0.189739719,
-0.9999361038,
0.3270712793,
0.4043011963,
-0.6823720932,
-0.3455865085,
-0.3148141205,
0.2707005143,
-0.6510784626,
1.0802608728,
0.1271724999,
0.0133321593,
0.0838312209,
-0.03... |
http://arxiv.org/abs/2406.12018v1 | http://arxiv.org/pdf/2406.12018v1 | CItruS: Chunked Instruction-aware State Eviction for Long Sequence Modeling | 2024-06-17T18:34:58 | 2024-06-17T18:34:58 | Yu Bai, Xiyuan Zou, Heyan Huang, Sanxing Chen, Marc-Antoine Rondeau, Yang Gao, Jackie Chi Kit Cheung | cs.CL | cs.CL | Long sequence modeling has gained broad interest as large language models
(LLMs) continue to advance. Recent research has identified that a large portion
of hidden states within the key-value caches of Transformer models can be
discarded (also termed evicted) without affecting the perplexity performance in
generating l... | [
-0.0448055193,
-0.5999765992,
0.6106800437,
0.2847694159,
-0.2240029275,
0.1099008918,
-0.2545138001,
-0.2821232975,
0.2867916524,
0.5551149845,
-0.4018616974,
-0.1664347053,
-0.5642663836,
-0.090449512,
-0.8300830126,
0.7968656421,
-0.0044053094,
-0.5555856824,
-0.0945677683,
... |
http://arxiv.org/abs/2406.12009v1 | http://arxiv.org/pdf/2406.12009v1 | FinTruthQA: A Benchmark Dataset for Evaluating the Quality of Financial Information Disclosure | 2024-06-17T18:25:02 | 2024-06-17T18:25:02 | Ziyue Xu, Peilin Zhou, Xinyu Shi, Jiageng Wu, Yikang Jiang, Bin Ke, Jie Yang | cs.CL | cs.CL | Accurate and transparent financial information disclosure is crucial in the
fields of accounting and finance, ensuring market efficiency and investor
confidence. Among many information disclosure platforms, the Chinese stock
exchanges' investor interactive platform provides a novel and interactive way
for listed firms ... | [
-0.2731298804,
-0.5722950101,
-0.0374803692,
0.43415609,
-0.235532105,
0.2381607592,
-0.0981555656,
-0.7527549863,
-0.1424339265,
0.387403667,
-0.5232319236,
-0.5108267665,
-0.3132976294,
-0.0559169054,
-0.4371013343,
0.7927997708,
-0.0679288954,
0.185069859,
-0.1692732573,
-0.... |
http://arxiv.org/abs/2406.11978v1 | http://arxiv.org/pdf/2406.11978v1 | Dialogue Action Tokens: Steering Language Models in Goal-Directed Dialogue with a Multi-Turn Planner | 2024-06-17T18:01:32 | 2024-06-17T18:01:32 | Kenneth Li, Yiming Wang, Fernanda Viégas, Martin Wattenberg | cs.CL | cs.CL, cs.AI, cs.LG | We present an approach called Dialogue Action Tokens (DAT) that adapts
language model agents to plan goal-directed dialogues. The core idea is to
treat each utterance as an action, thereby converting dialogues into games
where existing approaches such as reinforcement learning can be applied.
Specifically, we freeze a ... | [
-0.1133089215,
-0.9664006829,
0.5734096766,
0.3035818338,
-0.2796799839,
0.3732936382,
-0.1170514673,
-0.0087842494,
0.5400233865,
0.7552266717,
-0.893765986,
-0.5870253444,
-0.5148807764,
-0.0508456565,
-0.3091432452,
1.0259916782,
-0.0207854174,
0.3218994439,
0.2240179628,
-0... |
http://arxiv.org/abs/2406.11977v1 | http://arxiv.org/pdf/2406.11977v1 | Reframing linguistic bootstrapping as joint inference using visually-grounded grammar induction models | 2024-06-17T18:01:06 | 2024-06-17T18:01:06 | Eva Portelance, Siva Reddy, Timothy J. O'Donnell | cs.CL | cs.CL, I.2.7; I.2.10; I.2.6; F.4.2 | Semantic and syntactic bootstrapping posit that children use their prior
knowledge of one linguistic domain, say syntactic relations, to help later
acquire another, such as the meanings of new words. Empirical results
supporting both theories may tempt us to believe that these are different
learning strategies, where o... | [
-0.0264175385,
-0.8233875036,
0.6866919994,
0.0741858929,
-0.2308457643,
0.2119642347,
-0.3873152137,
-0.5776270032,
0.0312623568,
0.1176620796,
-0.5381903052,
-0.2133048773,
-0.6712164283,
-0.0889320076,
-0.3502328098,
0.621137917,
-0.0808120295,
0.3030213118,
-0.6397861838,
-... |
http://arxiv.org/abs/2406.11839v1 | http://arxiv.org/pdf/2406.11839v1 | mDPO: Conditional Preference Optimization for Multimodal Large Language Models | 2024-06-17T17:59:58 | 2024-06-17T17:59:58 | Fei Wang, Wenxuan Zhou, James Y. Huang, Nan Xu, Sheng Zhang, Hoifung Poon, Muhao Chen | cs.CV | cs.CV, cs.AI, cs.CL, cs.LG | Direct preference optimization (DPO) has shown to be an effective method for
large language model (LLM) alignment. Recent works have attempted to apply DPO
to multimodal scenarios but have found it challenging to achieve consistent
improvement. Through a comparative experiment, we identify the unconditional
preference ... | [
-0.4903047979,
-0.4870889187,
0.8844309449,
0.3037443459,
-0.3574803174,
-0.006051498,
-0.4072723389,
-0.2070407718,
0.229097113,
0.4327497482,
-0.6129021645,
-0.2320208699,
-0.7363587618,
0.1004857719,
0.232040152,
0.4802645147,
-0.0810311511,
0.0930886343,
-0.5377064943,
-0.1... |
http://arxiv.org/abs/2406.11830v1 | http://arxiv.org/pdf/2406.11830v1 | Language Modeling with Editable External Knowledge | 2024-06-17T17:59:35 | 2024-06-17T17:59:35 | Belinda Z. Li, Emmy Liu, Alexis Ross, Abbas Zeitoun, Graham Neubig, Jacob Andreas | cs.CL | cs.CL, cs.AI | When the world changes, so does the text that humans write about it. How do
we build language models that can be easily updated to reflect these changes?
One popular approach is retrieval-augmented generation, in which new documents
are inserted into a knowledge base and retrieved during prediction for
downstream tasks... | [
-0.3598380387,
-1.0424029827,
0.5122615099,
-0.1574935019,
0.2212003171,
0.2185733616,
-0.2814507186,
-0.7326890826,
0.0692351758,
0.6177255511,
-0.3515013158,
-0.1571499407,
-0.4931360185,
-0.0962782875,
-0.7531453967,
0.9063711762,
0.0154767195,
-0.3917945027,
-0.4184556603,
... |
http://arxiv.org/abs/2406.11827v1 | http://arxiv.org/pdf/2406.11827v1 | WPO: Enhancing RLHF with Weighted Preference Optimization | 2024-06-17T17:59:13 | 2024-06-17T17:59:13 | Wenxuan Zhou, Ravi Agrawal, Shujian Zhang, Sathish Reddy Indurthi, Sanqiang Zhao, Kaiqiang Song, Silei Xu, Chenguang Zhu | cs.CL | cs.CL, cs.AI, cs.LG | Reinforcement learning from human feedback (RLHF) is a promising solution to
align large language models (LLMs) more closely with human values. Off-policy
preference optimization, where the preference data is obtained from other
models, is widely adopted due to its cost efficiency and scalability. However,
off-policy p... | [
-0.4785722196,
-0.6221610904,
0.3261116445,
0.164628759,
-0.1962518543,
-0.0482906364,
-0.4273785055,
-0.2522054315,
0.299664259,
0.4578850865,
-0.7171791792,
0.0121620698,
-0.4463918209,
-0.1037673503,
0.5294745564,
0.8951499462,
0.0532910042,
-0.004684899,
-0.2696548998,
-0.2... |
http://arxiv.org/abs/2406.11823v1 | http://arxiv.org/pdf/2406.11823v1 | On Efficient Language and Vision Assistants for Visually-Situated Natural Language Understanding: What Matters in Reading and Reasoning | 2024-06-17T17:57:30 | 2024-06-17T17:57:30 | Geewook Kim, Minjoon Seo | cs.CV | cs.CV, cs.CL | Recent advancements in language and vision assistants have showcased
impressive capabilities but suffer from a lack of transparency, limiting
broader research and reproducibility. While open-source models handle general
image tasks effectively, they face challenges with the high computational
demands of complex visuall... | [
-0.3343245089,
-0.7828989625,
0.7317690849,
0.1428331435,
-0.1788650304,
-0.4217170477,
-0.1765421778,
-0.8807255626,
-0.2160582989,
0.6916503906,
-0.2531569004,
-0.4110023975,
-0.3972657323,
0.0441649072,
-0.2379053682,
0.3672687113,
-0.1019203067,
-0.0509970896,
-0.436771512,
... |
http://arxiv.org/abs/2406.11817v1 | http://arxiv.org/pdf/2406.11817v1 | Iterative Length-Regularized Direct Preference Optimization: A Case Study on Improving 7B Language Models to GPT-4 Level | 2024-06-17T17:55:38 | 2024-06-17T17:55:38 | Jie Liu, Zhanhui Zhou, Jiaheng Liu, Xingyuan Bu, Chao Yang, Han-Sen Zhong, Wanli Ouyang | cs.CL | cs.CL, cs.AI, cs.LG | Direct Preference Optimization (DPO), a standard method for aligning language
models with human preferences, is traditionally applied to offline preferences.
Recent studies show that DPO benefits from iterative training with online
preferences labeled by a trained reward model. In this work, we identify a
pitfall of va... | [
-0.5024197102,
-0.5372958779,
0.5965969563,
0.2714615464,
-0.5068619847,
0.1008839533,
-0.5138687491,
-0.4407788217,
0.1471280605,
0.3431422114,
-0.3066658378,
-0.2738823593,
-0.6721900702,
0.3916147649,
0.1518636793,
0.7339923978,
-0.0996401012,
-0.074196443,
-0.0159733985,
0.... |
http://arxiv.org/abs/2406.11813v1 | http://arxiv.org/pdf/2406.11813v1 | How Do Large Language Models Acquire Factual Knowledge During Pretraining? | 2024-06-17T17:54:40 | 2024-06-17T17:54:40 | Hoyeon Chang, Jinho Park, Seonghyeon Ye, Sohee Yang, Youngkyung Seo, Du-Seong Chang, Minjoon Seo | cs.CL | cs.CL, I.2.7 | Despite the recent observation that large language models (LLMs) can store
substantial factual knowledge, there is a limited understanding of the
mechanisms of how they acquire factual knowledge through pretraining. This work
addresses this gap by studying how LLMs acquire factual knowledge during
pretraining. The find... | [
-0.2121628076,
-0.7334625721,
1.1433297396,
-0.297403127,
-0.2683446109,
0.0709572434,
-0.4163078368,
-0.3751306236,
-0.1421407908,
0.4586906731,
-0.4657430053,
-0.2206623107,
-0.8662829399,
-0.1930393875,
-0.2766146958,
1.2256603241,
0.0375459194,
0.0291926842,
-0.4664913416,
... |
http://arxiv.org/abs/2406.11811v1 | http://arxiv.org/pdf/2406.11811v1 | RepLiQA: A Question-Answering Dataset for Benchmarking LLMs on Unseen Reference Content | 2024-06-17T17:52:54 | 2024-06-17T17:52:54 | Joao Monteiro, Pierre-Andre Noel, Etienne Marcotte, Sai Rajeswar, Valentina Zantedeschi, David Vazquez, Nicolas Chapados, Christopher Pal, Perouz Taslakian | cs.CL | cs.CL, cs.AI | Large Language Models (LLMs) are trained on vast amounts of data, most of
which is automatically scraped from the internet. This data includes
encyclopedic documents that harbor a vast amount of general knowledge (e.g.,
Wikipedia) but also potentially overlap with benchmark datasets used for
evaluating LLMs. Consequent... | [
-0.5677736998,
-0.5149022937,
0.4572927058,
0.0364650264,
-0.1686884314,
0.0705473274,
-0.0853395686,
-0.2715080976,
-0.0932075307,
0.7341943383,
-0.8192391396,
-0.266752094,
-0.3515782952,
0.2925597727,
-0.7189853787,
0.6968785524,
-0.1230938286,
0.1505329162,
-0.6736388206,
-... |
http://arxiv.org/abs/2406.11944v1 | http://arxiv.org/pdf/2406.11944v1 | Transcoders Find Interpretable LLM Feature Circuits | 2024-06-17T17:49:00 | 2024-06-17T17:49:00 | Jacob Dunefsky, Philippe Chlenski, Neel Nanda | cs.LG | cs.LG, cs.CL | A key goal in mechanistic interpretability is circuit analysis: finding
sparse subgraphs of models corresponding to specific behaviors or capabilities.
However, MLP sublayers make fine-grained circuit analysis on transformer-based
language models difficult. In particular, interpretable features -- such as
those found b... | [
-0.4046167135,
-0.4229216576,
0.5060725212,
0.2480870038,
-0.0250365417,
-0.0675597265,
-0.0510971397,
-0.4451335371,
0.2048555315,
0.5590550303,
-0.5950505733,
0.1093814149,
-0.842212081,
-0.0197724868,
-0.7357378006,
0.8316630721,
-0.2147332579,
0.0498063862,
0.1511365324,
-0... |
http://arxiv.org/abs/2406.11801v1 | http://arxiv.org/pdf/2406.11801v1 | Safety Arithmetic: A Framework for Test-time Safety Alignment of Language Models by Steering Parameters and Activations | 2024-06-17T17:48:13 | 2024-06-17T17:48:13 | Rima Hazra, Sayan Layek, Somnath Banerjee, Soujanya Poria | cs.CL | cs.CL | Ensuring the safe alignment of large language models (LLMs) with human values
is critical as they become integral to applications like translation and
question answering. Current alignment methods struggle with dynamic user
intentions and complex objectives, making models vulnerable to generating
harmful content. We pr... | [
0.0684423447,
-0.6568191648,
0.2552633882,
0.1931365281,
-0.125244841,
-0.0882829055,
0.1946834922,
-0.4449265599,
-0.3906112313,
0.5951806307,
-0.3734949529,
-0.5432673693,
-0.4842749834,
-0.2143238783,
-0.6918241978,
1.0463927984,
-0.0558574274,
0.0223806519,
-0.3043298721,
-... |
http://arxiv.org/abs/2406.11794v3 | http://arxiv.org/pdf/2406.11794v3 | DataComp-LM: In search of the next generation of training sets for language models | 2024-06-17T17:42:57 | 2024-06-20T17:43:05 | Jeffrey Li, Alex Fang, Georgios Smyrnis, Maor Ivgi, Matt Jordan, Samir Gadre, Hritik Bansal, Etash Guha, Sedrick Keh, Kushal Arora, Saurabh Garg, Rui Xin, Niklas Muennighoff, Reinhard Heckel, Jean Mercat, Mayee Chen, Suchin Gururangan, Mitchell Wortsman, Alon Albalak, Yonatan Bitton, Marianna Nezhurina, Amro Abbas, Che... | cs.LG | cs.LG, cs.CL | We introduce DataComp for Language Models (DCLM), a testbed for controlled
dataset experiments with the goal of improving language models. As part of
DCLM, we provide a standardized corpus of 240T tokens extracted from Common
Crawl, effective pretraining recipes based on the OpenLM framework, and a broad
suite of 53 do... | [
-0.3840751052,
-0.7096797228,
0.3048983514,
-0.1333581805,
-0.1820856184,
0.3032336831,
-0.3192958534,
-0.1119647473,
-0.36577034,
0.5960382819,
-0.5701571107,
-0.7432522178,
-0.573243618,
0.0755413026,
-0.6639484167,
0.8260958791,
-0.0477698371,
0.1636057645,
-0.1820099503,
-0... |
http://arxiv.org/abs/2406.11785v1 | http://arxiv.org/pdf/2406.11785v1 | CELL your Model: Contrastive Explanation Methods for Large Language Models | 2024-06-17T17:39:10 | 2024-06-17T17:39:10 | Ronny Luss, Erik Miehling, Amit Dhurandhar | cs.CL | cs.CL, cs.AI, cs.LG | The advent of black-box deep neural network classification models has sparked
the need to explain their decisions. However, in the case of generative AI such
as large language models (LLMs), there is no class prediction to explain.
Rather, one can ask why an LLM output a particular response to a given prompt.
In this p... | [
-0.5120928884,
-0.7962343097,
0.6318552494,
0.1636485159,
-0.2309836596,
-0.0061006914,
-0.3001076877,
-0.4507491887,
0.0958796069,
0.3123490214,
-0.7391909361,
-0.4369622767,
-0.233989656,
0.0937945172,
-0.5065306425,
0.8967173696,
0.1575887501,
0.0743597373,
-0.4953229725,
-0... |
http://arxiv.org/abs/2406.11784v1 | http://arxiv.org/pdf/2406.11784v1 | MDCR: A Dataset for Multi-Document Conditional Reasoning | 2024-06-17T17:38:43 | 2024-06-17T17:38:43 | Peter Baile Chen, Yi Zhang, Chunwei Liu, Sejal Gupta, Yoon Kim, Michael Cafarella | cs.CL | cs.CL, cs.AI | The same real-life questions posed to different individuals may lead to
different answers based on their unique situations. For instance, whether a
student is eligible for a scholarship depends on eligibility conditions, such
as major or degree required. ConditionalQA was proposed to evaluate models'
capability of read... | [
-0.4632113874,
-0.6092410684,
0.9120678306,
-0.1756113172,
-0.0104531012,
0.2000395656,
-0.2171052396,
-0.033086244,
-0.180063352,
0.5292788744,
-1.0661714077,
-0.2642823458,
-0.6493319869,
0.425385654,
-0.3766458929,
0.8043872118,
-0.4417660534,
0.3941149712,
-0.7294345498,
-0... |
http://arxiv.org/abs/2406.11780v1 | http://arxiv.org/pdf/2406.11780v1 | Split, Unlearn, Merge: Leveraging Data Attributes for More Effective Unlearning in LLMs | 2024-06-17T17:35:52 | 2024-06-17T17:35:52 | Swanand Ravindra Kadhe, Farhan Ahmed, Dennis Wei, Nathalie Baracaldo, Inkit Padhi | cs.LG | cs.LG, cs.AI, cs.CL | Large language models (LLMs) have shown to pose social and ethical risks such
as generating toxic language or facilitating malicious use of hazardous
knowledge. Machine unlearning is a promising approach to improve LLM safety by
directly removing harmful behaviors and knowledge. In this paper, we propose
"SPlit, UNlear... | [
-0.2919432223,
-0.6021078825,
0.3388310969,
-0.3637621999,
-0.2572574914,
0.6100685596,
-0.4785698056,
-0.4192431569,
-0.2355304658,
0.679728806,
-0.57822299,
-0.4869350791,
-0.6244575381,
-0.4310823381,
-0.7615880966,
1.1944384575,
0.3370776772,
0.2171418518,
-0.4530396163,
-0... |
http://arxiv.org/abs/2406.11776v1 | http://arxiv.org/pdf/2406.11776v1 | Improving Multi-Agent Debate with Sparse Communication Topology | 2024-06-17T17:33:09 | 2024-06-17T17:33:09 | Yunxuan Li, Yibing Du, Jiageng Zhang, Le Hou, Peter Grabowski, Yeqing Li, Eugene Ie | cs.CL | cs.CL | Multi-agent debate has proven effective in improving large language models
quality for reasoning and factuality tasks. While various role-playing
strategies in multi-agent debates have been explored, in terms of the
communication among agents, existing approaches adopt a brute force algorithm
-- each agent can communic... | [
-0.8924534917,
-1.0454614162,
0.7721531987,
-0.121754773,
-0.1174944788,
0.0540929101,
-0.2027864009,
-0.4688165188,
0.4693554342,
0.318844676,
-0.4583507776,
-0.1452849805,
-0.5896331668,
0.103903234,
-0.0490775667,
0.706037879,
0.3839853108,
-0.1470184624,
-0.3927140832,
-0.5... |
http://arxiv.org/abs/2406.11768v1 | http://arxiv.org/pdf/2406.11768v1 | GAMA: A Large Audio-Language Model with Advanced Audio Understanding and Complex Reasoning Abilities | 2024-06-17T17:31:01 | 2024-06-17T17:31:01 | Sreyan Ghosh, Sonal Kumar, Ashish Seth, Chandra Kiran Reddy Evuru, Utkarsh Tyagi, S Sakshi, Oriol Nieto, Ramani Duraiswami, Dinesh Manocha | cs.SD | cs.SD, cs.AI, cs.CL, eess.AS | Perceiving and understanding non-speech sounds and non-verbal speech is
essential to making decisions that help us interact with our surroundings. In
this paper, we propose GAMA, a novel General-purpose Large Audio-Language Model
(LALM) with Advanced Audio Understanding and Complex Reasoning Abilities. We
build GAMA by... | [
-0.6350906491,
-1.1902170181,
0.7881093025,
0.0534220263,
0.0391016304,
0.1405159384,
-0.4396966398,
-0.5393177867,
-0.1489755809,
0.3139323592,
-0.6280822754,
-0.6442725658,
-0.5547194481,
-0.2855234444,
-0.4805061221,
0.5184258223,
0.3589403927,
0.0934784263,
-0.1236279085,
-... |
http://arxiv.org/abs/2406.11939v1 | http://arxiv.org/pdf/2406.11939v1 | From Crowdsourced Data to High-Quality Benchmarks: Arena-Hard and BenchBuilder Pipeline | 2024-06-17T17:26:10 | 2024-06-17T17:26:10 | Tianle Li, Wei-Lin Chiang, Evan Frick, Lisa Dunlap, Tianhao Wu, Banghua Zhu, Joseph E. Gonzalez, Ion Stoica | cs.LG | cs.LG, cs.AI, cs.CL | The rapid evolution of language models has necessitated the development of
more challenging benchmarks. Current static benchmarks often struggle to
consistently distinguish between the capabilities of different models and fail
to align with real-world user preferences. On the other hand, live
crowd-sourced platforms li... | [
-0.4941114187,
-0.5170242786,
0.3669779599,
0.3304568827,
0.0713718459,
0.3489997685,
-0.3770750761,
-0.3397717476,
-0.2379645854,
0.4660158455,
-0.5256636739,
-0.4216904044,
-0.3880045414,
0.1541215479,
-0.2230223119,
0.7450481057,
-0.1023896113,
-0.0578480549,
-0.6433695555,
... |
http://arxiv.org/abs/2406.11757v1 | http://arxiv.org/pdf/2406.11757v1 | STAR: SocioTechnical Approach to Red Teaming Language Models | 2024-06-17T17:16:45 | 2024-06-17T17:16:45 | Laura Weidinger, John Mellor, Bernat Guillen Pegueroles, Nahema Marchal, Ravin Kumar, Kristian Lum, Canfer Akbulut, Mark Diaz, Stevie Bergman, Mikel Rodriguez, Verena Rieser, William Isaac | cs.AI | cs.AI, cs.CL, cs.CY, cs.HC | This research introduces STAR, a sociotechnical framework that improves on
current best practices for red teaming safety of large language models. STAR
makes two key contributions: it enhances steerability by generating
parameterised instructions for human red teamers, leading to improved coverage
of the risk surface. ... | [
-0.2131556869,
-0.6980269551,
0.2499888539,
0.2488868088,
0.1842713058,
0.4400933981,
-0.2916768491,
-0.9504737854,
0.4221822917,
0.4850458503,
-0.3884240687,
-0.5427013636,
-0.6960834265,
-0.086245738,
-0.2022433579,
1.1402155161,
0.1906913221,
0.0853010193,
0.1689182967,
0.01... |
http://arxiv.org/abs/2406.11753v1 | http://arxiv.org/pdf/2406.11753v1 | A Semantic-based Layer Freezing Approach to Efficient Fine-Tuning of Language Models | 2024-06-17T17:13:08 | 2024-06-17T17:13:08 | Jian Gu, Aldeida Aleti, Chunyang Chen, Hongyu Zhang | cs.CL | cs.CL, cs.LG | Finetuning language models (LMs) is crucial for adapting the models to
downstream data and tasks. However, full finetuning is usually costly. Existing
work, such as parameter-efficient finetuning (PEFT), often focuses on
\textit{how to finetune} but neglects the issue of \textit{where to finetune}.
As a pioneering work... | [
-0.650601387,
-1.1960084438,
0.6688848138,
-0.1667728722,
-0.1237311289,
-0.1655090898,
-0.4462553561,
-0.2651974261,
-0.299051106,
0.7888857126,
-0.7373004556,
-0.5539466143,
-0.2833899856,
-0.3282865286,
-0.3615957201,
0.75956738,
0.1778634936,
-0.1460901499,
-0.1052498147,
0... |
http://arxiv.org/abs/2406.11745v1 | http://arxiv.org/pdf/2406.11745v1 | Multi-Layer Ranking with Large Language Models for News Source Recommendation | 2024-06-17T17:02:34 | 2024-06-17T17:02:34 | Wenjia Zhang, Lin Gui, Rob Procter, Yulan He | cs.IR | cs.IR, cs.CL | To seek reliable information sources for news events, we introduce a novel
task of expert recommendation, which aims to identify trustworthy sources based
on their previously quoted statements. To achieve this, we built a novel
dataset, called NewsQuote, consisting of 23,571 quote-speaker pairs sourced
from a collectio... | [
-0.3029673398,
-0.5807110667,
0.7735335827,
-0.0381612852,
-0.3826580644,
0.2817368507,
-0.1308030933,
-0.7363225222,
0.2278998941,
0.5972335339,
-0.2917089164,
-0.5894265771,
-0.6577033997,
-0.3177343309,
-0.3370819092,
1.0476256609,
0.2742449641,
0.1993652433,
-0.2138813138,
... |
http://arxiv.org/abs/2406.11736v1 | http://arxiv.org/pdf/2406.11736v1 | Interactive Evolution: A Neural-Symbolic Self-Training Framework For Large Language Models | 2024-06-17T16:52:56 | 2024-06-17T16:52:56 | Fangzhi Xu, Qiushi Sun, Kanzhi Cheng, Jun Liu, Yu Qiao, Zhiyong Wu | cs.CL | cs.CL, cs.AI | One of the primary driving forces contributing to the superior performance of
Large Language Models (LLMs) is the extensive availability of human-annotated
natural language data, which is used for alignment fine-tuning. This inspired
researchers to investigate self-training methods to mitigate the extensive
reliance on... | [
-0.2432887256,
-0.7095466852,
0.5827829242,
0.393907398,
-0.257773906,
0.154384017,
-0.4299070239,
-0.4191028178,
0.3691155016,
0.3561253846,
-0.8121209741,
-0.2899178267,
-0.5300469995,
0.4433603883,
-0.3154861629,
0.6132524014,
-0.1651527733,
0.0072675841,
-0.3573450148,
-0.0... |
http://arxiv.org/abs/2406.11727v2 | http://arxiv.org/pdf/2406.11727v2 | 1000 African Voices: Advancing inclusive multi-speaker multi-accent speech synthesis | 2024-06-17T16:46:10 | 2024-06-27T08:52:54 | Sewade Ogun, Abraham T. Owodunni, Tobi Olatunji, Eniola Alese, Babatunde Oladimeji, Tejumade Afonja, Kayode Olaleye, Naome A. Etori, Tosin Adewumi | eess.AS | eess.AS, cs.CL | Recent advances in speech synthesis have enabled many useful applications
like audio directions in Google Maps, screen readers, and automated content
generation on platforms like TikTok. However, these systems are mostly
dominated by voices sourced from data-rich geographies with personas
representative of their source... | [
-1.0412346125,
-0.3221873343,
0.3177231848,
0.5631875992,
-0.1040797159,
0.5302410126,
-0.2248375714,
-0.4846705198,
0.8754921556,
0.2850693166,
-0.6740638614,
-0.3298178017,
-0.3906323314,
0.5081533194,
-0.3906896114,
0.5513996482,
0.7967854142,
0.0056531839,
0.4460238218,
-0.... |
http://arxiv.org/abs/2406.11721v1 | http://arxiv.org/pdf/2406.11721v1 | Zero-Shot Generalization during Instruction Tuning: Insights from Similarity and Granularity | 2024-06-17T16:40:21 | 2024-06-17T16:40:21 | Bingxiang He, Ning Ding, Cheng Qian, Jia Deng, Ganqu Cui, Lifan Yuan, Huan-ang Gao, Huimin Chen, Zhiyuan Liu, Maosong Sun | cs.CL | cs.CL, cs.AI, cs.LG | Understanding alignment techniques begins with comprehending zero-shot
generalization brought by instruction tuning, but little of the mechanism has
been understood. Existing work has largely been confined to the task level,
without considering that tasks are artificially defined and, to LLMs, merely
consist of tokens ... | [
-0.2222274691,
-1.0482873917,
0.6071138978,
-0.2881088555,
-0.0838148966,
-0.1276324838,
-0.190464139,
-0.2437072992,
-0.0565525293,
0.2689108253,
-0.7150447965,
-0.7019093037,
-0.3936781585,
-0.4429862201,
-0.524645865,
1.2160630226,
-0.0953761339,
-0.0601152219,
-0.4464695156,
... |
http://arxiv.org/abs/2406.11717v1 | http://arxiv.org/pdf/2406.11717v1 | Refusal in Language Models Is Mediated by a Single Direction | 2024-06-17T16:36:12 | 2024-06-17T16:36:12 | Andy Arditi, Oscar Obeso, Aaquib Syed, Daniel Paleka, Nina Rimsky, Wes Gurnee, Neel Nanda | cs.LG | cs.LG, cs.AI, cs.CL | Conversational large language models are fine-tuned for both
instruction-following and safety, resulting in models that obey benign requests
but refuse harmful ones. While this refusal behavior is widespread across chat
models, its underlying mechanisms remain poorly understood. In this work, we
show that refusal is me... | [
-0.3779267669,
-1.1690444946,
0.4894081652,
0.141884461,
-0.1186886728,
-0.0133740455,
-0.2147176266,
0.0534490608,
0.1437625289,
0.8290019631,
-1.001034379,
-0.3279141188,
-0.3973026872,
-0.3558811843,
-0.4945092499,
1.1358908415,
0.2162833959,
-0.1149510965,
-0.1997112781,
-0... |
http://arxiv.org/abs/2406.11715v1 | http://arxiv.org/pdf/2406.11715v1 | Measuring memorization in RLHF for code completion | 2024-06-17T16:33:35 | 2024-06-17T16:33:35 | Aneesh Pappu, Billy Porter, Ilia Shumailov, Jamie Hayes | cs.LG | cs.LG, cs.CL, cs.SE | Reinforcement learning with human feedback (RLHF) has become the dominant
method to align large models to user preferences. Unlike fine-tuning, for which
there are many studies regarding training data memorization, it is not clear
how memorization is affected by or introduced in the RLHF alignment process.
Understandin... | [
-0.3668406606,
-0.5820839405,
0.3243516088,
-0.0334145948,
-0.0950456783,
0.1123621166,
-0.4199343622,
-0.4625627995,
-0.2252242714,
0.6229289174,
-0.6472937465,
-0.5663406849,
-0.1189861,
-0.086527206,
-0.3019218743,
1.4033913612,
-0.3609051108,
-0.1196922511,
-0.4991173744,
-... |
http://arxiv.org/abs/2406.11709v2 | http://arxiv.org/pdf/2406.11709v2 | Instruct, Not Assist: LLM-based Multi-Turn Planning and Hierarchical Questioning for Socratic Code Debugging | 2024-06-17T16:28:21 | 2024-06-25T21:17:41 | Priyanka Kargupta, Ishika Agarwal, Dilek Hakkani-Tur, Jiawei Han | cs.CL | cs.CL, cs.MA | Socratic questioning is an effective teaching strategy, encouraging critical
thinking and problem-solving. The conversational capabilities of large language
models (LLMs) show great potential for providing scalable, real-time student
guidance. However, current LLMs often give away solutions directly, making them
ineffe... | [
-0.0969098657,
-0.8766265512,
0.5131642222,
0.232657671,
0.0188883692,
0.7955908775,
-0.0320281573,
-0.3363418579,
0.106630519,
0.2777511179,
-0.5113001466,
-0.6867184043,
-0.2592860758,
-0.1693819314,
-0.4800154567,
0.9654114246,
-0.0287791416,
-0.0029635034,
-0.5118188858,
0.... |
http://arxiv.org/abs/2406.11706v1 | http://arxiv.org/pdf/2406.11706v1 | Prompts as Auto-Optimized Training Hyperparameters: Training Best-in-Class IR Models from Scratch with 10 Gold Labels | 2024-06-17T16:25:55 | 2024-06-17T16:25:55 | Jasper Xian, Saron Samuel, Faraz Khoubsirat, Ronak Pradeep, Md Arafat Sultan, Radu Florian, Salim Roukos, Avirup Sil, Christopher Potts, Omar Khattab | cs.IR | cs.IR, cs.CL, cs.LG | We develop a method for training small-scale (under 100M parameter) neural
information retrieval models with as few as 10 gold relevance labels. The
method depends on generating synthetic queries for documents using a language
model (LM), and the key step is that we automatically optimize the LM prompt
that is used to ... | [
-0.5489569902,
-0.4867022038,
0.6049937606,
-0.0897466838,
0.0081182355,
-0.1321291476,
-0.1233734936,
-0.2336619496,
-0.0856030062,
0.4096575975,
-0.5527185798,
-0.2411682755,
-0.3990690708,
0.4885649383,
-0.201988712,
1.2090123892,
0.3260711133,
0.2853472233,
-0.1675588787,
-... |
http://arxiv.org/abs/2406.11704v1 | http://arxiv.org/pdf/2406.11704v1 | Nemotron-4 340B Technical Report | 2024-06-17T16:25:04 | 2024-06-17T16:25:04 | Nvidia, :, Bo Adler, Niket Agarwal, Ashwath Aithal, Dong H. Anh, Pallab Bhattacharya, Annika Brundyn, Jared Casper, Bryan Catanzaro, Sharon Clay, Jonathan Cohen, Sirshak Das, Ayush Dattagupta, Olivier Delalleau, Leon Derczynski, Yi Dong, Daniel Egert, Ellie Evans, Aleksander Ficek, Denys Fridman, Shaona Ghosh, Boris ... | cs.CL | cs.CL, cs.AI, cs.LG | We release the Nemotron-4 340B model family, including Nemotron-4-340B-Base,
Nemotron-4-340B-Instruct, and Nemotron-4-340B-Reward. Our models are open
access under the NVIDIA Open Model License Agreement, a permissive model
license that allows distribution, modification, and use of the models and its
outputs. These mod... | [
-0.4830032289,
-0.2444198728,
0.4208781421,
0.1015999615,
-0.0601777136,
-0.2363303155,
0.1809290349,
-0.5366506577,
-0.1600935608,
0.8153895736,
-0.5507670641,
-0.2242526561,
-0.3337628543,
0.1451750249,
-0.2330077887,
0.93032974,
-0.2128972262,
0.1944788098,
-0.2137058079,
-0... |
http://arxiv.org/abs/2406.11698v1 | http://arxiv.org/pdf/2406.11698v1 | Meta Reasoning for Large Language Models | 2024-06-17T16:14:11 | 2024-06-17T16:14:11 | Peizhong Gao, Ao Xie, Shaoguang Mao, Wenshan Wu, Yan Xia, Haipeng Mi, Furu Wei | cs.CL | cs.CL | We introduce Meta-Reasoning Prompting (MRP), a novel and efficient system
prompting method for large language models (LLMs) inspired by human
meta-reasoning. Traditional in-context learning-based reasoning techniques,
such as Tree-of-Thoughts, show promise but lack consistent state-of-the-art
performance across diverse... | [
-0.4780264795,
-0.7364791036,
0.7222219706,
0.2338746786,
-0.0216486454,
0.1518512666,
-0.3850823641,
-0.2298161536,
-0.0789910033,
0.2703429461,
-0.9313655496,
-0.4314589798,
-0.2525688112,
0.419157654,
0.0867921114,
0.4853150547,
0.1444767416,
-0.1739200205,
-0.3996472359,
0.... |
http://arxiv.org/abs/2406.11695v1 | http://arxiv.org/pdf/2406.11695v1 | Optimizing Instructions and Demonstrations for Multi-Stage Language Model Programs | 2024-06-17T16:12:03 | 2024-06-17T16:12:03 | Krista Opsahl-Ong, Michael J Ryan, Josh Purtell, David Broman, Christopher Potts, Matei Zaharia, Omar Khattab | cs.CL | cs.CL, cs.AI, cs.LG | Language Model Programs, i.e. sophisticated pipelines of modular language
model (LM) calls, are increasingly advancing NLP tasks, but they require
crafting prompts that are jointly effective for all modules. We study prompt
optimization for LM programs, i.e. how to update these prompts to maximize a
downstream metric w... | [
-0.2284344435,
-0.7347896099,
0.6672330499,
0.4476107955,
-0.0443655737,
0.0312234946,
-0.3023971021,
0.1185886264,
-0.2054282129,
0.4028812647,
-0.9205960035,
-0.7028137445,
-0.2733157873,
-0.0007429675,
-0.1198259443,
0.9878320098,
0.0854679644,
0.0013451354,
-0.3767230213,
-... |
http://arxiv.org/abs/2406.11687v1 | http://arxiv.org/pdf/2406.11687v1 | Tokenization Falling Short: The Curse of Tokenization | 2024-06-17T16:05:32 | 2024-06-17T16:05:32 | Yekun Chai, Yewei Fang, Qiwei Peng, Xuhong Li | cs.CL | cs.CL | Language models typically tokenize raw text into sequences of subword
identifiers from a predefined vocabulary, a process inherently sensitive to
typographical errors, length variations, and largely oblivious to the internal
structure of tokens-issues we term the curse of tokenization. In this study, we
delve into thes... | [
-0.1215056926,
-0.8028686047,
0.5615836978,
0.0793685988,
-0.4417783618,
0.2617834806,
-0.1469078362,
-0.5108292699,
0.1278814673,
0.5038545728,
-0.3435299098,
-0.0051754736,
-0.9294210672,
0.2196789533,
-0.337248683,
1.2663476467,
-0.1093469635,
0.1056055576,
-0.3155241907,
-0... |
http://arxiv.org/abs/2406.11683v1 | http://arxiv.org/pdf/2406.11683v1 | HoLLMwood: Unleashing the Creativity of Large Language Models in Screenwriting via Role Playing | 2024-06-17T16:01:33 | 2024-06-17T16:01:33 | Jing Chen, Xinyu Zhu, Cheng Yang, Chufan Shi, Yadong Xi, Yuxiang Zhang, Junjie Wang, Jiashu Pu, Rongsheng Zhang, Yujiu Yang, Tian Feng | cs.CL | cs.CL | Generative AI has demonstrated unprecedented creativity in the field of
computer vision, yet such phenomena have not been observed in natural language
processing. In particular, large language models (LLMs) can hardly produce
written works at the level of human experts due to the extremely high
complexity of literature... | [
-0.1157223135,
-0.9115809202,
0.7735998631,
0.145801574,
-0.0849960223,
0.3183170259,
-0.1453825831,
-0.1754072011,
0.0960152224,
1.0176705122,
-0.6426917315,
-0.6145902872,
-0.4408506453,
0.0046497257,
-0.2490903586,
0.8078087568,
-0.2427498102,
-0.2153097689,
-0.4125017524,
0... |
http://arxiv.org/abs/2406.11682v1 | http://arxiv.org/pdf/2406.11682v1 | Knowledge-to-Jailbreak: One Knowledge Point Worth One Attack | 2024-06-17T15:59:59 | 2024-06-17T15:59:59 | Shangqing Tu, Zhuoran Pan, Wenxuan Wang, Zhexin Zhang, Yuliang Sun, Jifan Yu, Hongning Wang, Lei Hou, Juanzi Li | cs.CL | cs.CL, cs.AI, cs.CR | Large language models (LLMs) have been increasingly applied to various
domains, which triggers increasing concerns about LLMs' safety on specialized
domains, e.g. medicine. However, testing the domain-specific safety of LLMs is
challenging due to the lack of domain knowledge-driven attacks in existing
benchmarks. To br... | [
-0.2682323158,
-0.8284702897,
0.6909354925,
0.0625394806,
-0.0558669865,
0.0844492838,
-0.3367516696,
-0.4702040553,
-0.000092302,
0.9647318721,
-0.3827127516,
-0.6926832795,
-0.7735257745,
-0.1770673543,
-0.5054125786,
1.091686368,
0.0103634065,
-0.2940903306,
-0.163612172,
0.... |
http://arxiv.org/abs/2406.11681v1 | http://arxiv.org/pdf/2406.11681v1 | R-Eval: A Unified Toolkit for Evaluating Domain Knowledge of Retrieval Augmented Large Language Models | 2024-06-17T15:59:49 | 2024-06-17T15:59:49 | Shangqing Tu, Yuanchun Wang, Jifan Yu, Yuyang Xie, Yaran Shi, Xiaozhi Wang, Jing Zhang, Lei Hou, Juanzi Li | cs.CL | cs.CL, cs.AI | Large language models have achieved remarkable success on general NLP tasks,
but they may fall short for domain-specific problems. Recently, various
Retrieval-Augmented Large Language Models (RALLMs) are proposed to address this
shortcoming. However, existing evaluation tools only provide a few baselines
and evaluate t... | [
-0.4175177217,
-0.876529336,
0.3640807271,
0.2336561233,
-0.0345083885,
0.4592424631,
-0.5682424903,
-0.3405152857,
0.0854261667,
0.7360100746,
-0.3158583343,
-0.5013812184,
-0.5898496509,
0.2635681033,
-0.3076678514,
1.1292434931,
-0.1656142175,
-0.1863303483,
-0.4867765307,
-... |
http://arxiv.org/abs/2406.11678v1 | http://arxiv.org/pdf/2406.11678v1 | TourRank: Utilizing Large Language Models for Documents Ranking with a Tournament-Inspired Strategy | 2024-06-17T15:58:22 | 2024-06-17T15:58:22 | Yiqun Chen, Qi Liu, Yi Zhang, Weiwei Sun, Daiting Shi, Jiaxin Mao, Dawei Yin | cs.IR | cs.IR, cs.CL | Large Language Models (LLMs) are increasingly employed in zero-shot documents
ranking, yielding commendable results. However, several significant challenges
still persist in LLMs for ranking: (1) LLMs are constrained by limited input
length, precluding them from processing a large number of documents
simultaneously; (2... | [
-0.2549721897,
-0.804905355,
0.6593680978,
0.0505949818,
-0.198606804,
0.4810726345,
-0.3993816376,
-0.3725474775,
0.2763235867,
0.3577073812,
-0.4276483059,
-0.8365299702,
-0.9457598329,
0.1651369929,
-0.4282376468,
0.8821766973,
0.122247681,
0.0135418847,
-0.2347650081,
-0.17... |
http://arxiv.org/abs/2406.11675v2 | http://arxiv.org/pdf/2406.11675v2 | BLoB: Bayesian Low-Rank Adaptation by Backpropagation for Large Language Models | 2024-06-17T15:55:38 | 2024-06-18T15:15:04 | Yibin Wang, Haizhou Shi, Ligong Han, Dimitris Metaxas, Hao Wang | cs.LG | cs.LG, cs.AI, cs.CL, stat.ML | Large Language Models (LLMs) often suffer from overconfidence during
inference, particularly when adapted to downstream domain-specific tasks with
limited data. Previous work addresses this issue by employing approximate
Bayesian estimation after the LLMs are trained, enabling them to quantify
uncertainty. However, suc... | [
-0.2185429037,
-0.8065010905,
0.6051664352,
0.1843708754,
-0.1511816233,
0.0762489736,
-0.5461207628,
-0.13715069,
-0.1321412772,
0.4421854913,
-0.4685071409,
-0.7467122674,
-1.0091633797,
-0.4007128477,
0.0662817806,
0.9276405573,
0.2548323572,
-0.000295447,
-0.0050635012,
-0.... |
http://arxiv.org/abs/2406.11674v1 | http://arxiv.org/pdf/2406.11674v1 | Endor: Hardware-Friendly Sparse Format for Offloaded LLM Inference | 2024-06-17T15:55:08 | 2024-06-17T15:55:08 | Donghyeon Joo, Ramyad Hadidi, Soheil Feizi, Bahar Asgari | cs.CL | cs.CL | The increasing size of large language models (LLMs) challenges their usage on
resource-constrained platforms. For example, memory on modern GPUs is
insufficient to hold LLMs that are hundreds of Gigabytes in size. Offloading is
a popular method to escape this constraint by storing weights of an LLM model
to host CPU me... | [
-0.2697323561,
-0.833057642,
0.4707916081,
0.1819898039,
-0.5766135454,
-0.1280873269,
-0.3146588802,
-0.2799058259,
0.3911189139,
0.5272156,
-0.3210826814,
0.0772747099,
-0.5190008283,
-0.1863146871,
-0.4598914087,
0.8808190227,
-0.1908307225,
-0.3195951283,
-0.5069041848,
-0.... |
http://arxiv.org/abs/2406.11670v1 | http://arxiv.org/pdf/2406.11670v1 | Benchmarking of LLM Detection: Comparing Two Competing Approaches | 2024-06-17T15:51:46 | 2024-06-17T15:51:46 | Thorsten Pröhl, Erik Putzier, Rüdiger Zarnekow | cs.CL | cs.CL, cs.AI | This article gives an overview of the field of LLM text recognition.
Different approaches and implemented detectors for the recognition of
LLM-generated text are presented. In addition to discussing the
implementations, the article focuses on benchmarking the detectors. Although
there are numerous software products for... | [
-0.2017240673,
-0.3610774875,
0.6688663363,
-0.0633704588,
-0.342572242,
0.2995778024,
-0.2000527978,
-0.7039430737,
-0.0612018742,
0.2647502422,
-0.3304947615,
-0.5569500327,
-0.8222583532,
0.441506654,
-0.4115299881,
0.6615071893,
-0.2721832097,
0.2451958358,
-0.2580899298,
-... |
http://arxiv.org/abs/2406.11668v1 | http://arxiv.org/pdf/2406.11668v1 | "Not Aligned" is Not "Malicious": Being Careful about Hallucinations of Large Language Models' Jailbreak | 2024-06-17T15:51:01 | 2024-06-17T15:51:01 | Lingrui Mei, Shenghua Liu, Yiwei Wang, Baolong Bi, Jiayi Mao, Xueqi Cheng | cs.CL | cs.CL | "Jailbreak" is a major safety concern of Large Language Models (LLMs), which
occurs when malicious prompts lead LLMs to produce harmful outputs, raising
issues about the reliability and safety of LLMs. Therefore, an effective
evaluation of jailbreaks is very crucial to develop its mitigation strategies.
However, our re... | [
-0.092040062,
-0.5924169421,
0.2155848742,
0.4570464492,
-0.2304162085,
0.2977258563,
-0.3770633936,
-0.8188117743,
0.4062474966,
0.5742256045,
-0.4577047527,
-0.4369439781,
-0.5521326661,
-0.0445871949,
-0.5458779335,
0.9841977358,
0.1338020116,
-0.1411149055,
-0.0851119384,
-... |
http://arxiv.org/abs/2406.11665v1 | http://arxiv.org/pdf/2406.11665v1 | See It from My Perspective: Diagnosing the Western Cultural Bias of Large Vision-Language Models in Image Understanding | 2024-06-17T15:49:51 | 2024-06-17T15:49:51 | Amith Ananthram, Elias Stengel-Eskin, Carl Vondrick, Mohit Bansal, Kathleen McKeown | cs.CL | cs.CL, cs.AI, cs.CV | Vision-language models (VLMs) can respond to queries about images in many
languages. However, beyond language, culture affects how we see things. For
example, individuals from Western cultures focus more on the central figure in
an image while individuals from Eastern cultures attend more to scene context.
In this work... | [
-0.6093233824,
-0.4270373881,
0.4066835344,
0.1786799282,
-0.3091754615,
-0.1638989002,
-0.4238762558,
-1.0020161867,
0.0571597517,
0.5153727531,
-0.6630570889,
-0.3323202431,
-0.6290113926,
0.0142963808,
-0.099253878,
0.6522186399,
-0.1019405052,
0.6486459374,
-0.1765020639,
-... |
http://arxiv.org/abs/2406.11661v2 | http://arxiv.org/pdf/2406.11661v2 | Cultural Conditioning or Placebo? On the Effectiveness of Socio-Demographic Prompting | 2024-06-17T15:43:45 | 2024-06-20T08:25:08 | Sagnik Mukherjee, Muhammad Farid Adilazuarda, Sunayana Sitaram, Kalika Bali, Alham Fikri Aji, Monojit Choudhury | cs.CL | cs.CL | Socio-demographic prompting is a commonly employed approach to study cultural
biases in LLMs as well as for aligning models to certain cultures. In this
paper, we systematically probe four LLMs (Llama 3, Mistral v0.2, GPT-3.5 Turbo
and GPT-4) with prompts that are conditioned on culturally sensitive and
non-sensitive c... | [
-0.2175343484,
-0.7158933878,
0.2986465096,
0.4417141974,
-0.2562722862,
0.1839413047,
-0.151937753,
-0.1784307808,
0.4506222308,
0.0600314178,
-0.5970376134,
-0.374462992,
-0.3528716266,
0.2942381501,
0.2301656008,
0.9590942264,
-0.146751821,
0.5549110174,
-0.017644424,
-0.160... |
http://arxiv.org/abs/2406.11657v1 | http://arxiv.org/pdf/2406.11657v1 | Can LLM be a Personalized Judge? | 2024-06-17T15:41:30 | 2024-06-17T15:41:30 | Yijiang River Dong, Tiancheng Hu, Nigel Collier | cs.CL | cs.CL, cs.CY | Ensuring that large language models (LLMs) reflect diverse user values and
preferences is crucial as their user bases expand globally. It is therefore
encouraging to see the growing interest in LLM personalization within the
research community. However, current works often rely on the LLM-as-a-Judge
approach for evalua... | [
-0.2829670012,
-0.5442960858,
1.1356011629,
0.321241349,
-0.0685188919,
0.2031005025,
-0.6568008065,
-0.4016728997,
0.211974889,
0.327113688,
-0.2260725051,
-0.7074648738,
-0.8324490786,
0.153827101,
-0.129355222,
0.6558988094,
0.2937494814,
0.2203522027,
-0.355761826,
-0.03775... |
http://arxiv.org/abs/2406.11654v1 | http://arxiv.org/pdf/2406.11654v1 | Ruby Teaming: Improving Quality Diversity Search with Memory for Automated Red Teaming | 2024-06-17T15:36:14 | 2024-06-17T15:36:14 | Vernon Toh Yan Han, Rishabh Bhardwaj, Soujanya Poria | cs.CL | cs.CL | We propose Ruby Teaming, a method that improves on Rainbow Teaming by
including a memory cache as its third dimension. The memory dimension provides
cues to the mutator to yield better-quality prompts, both in terms of attack
success rate (ASR) and quality diversity. The prompt archive generated by Ruby
Teaming has an ... | [
-0.6583871841,
-0.2419130653,
0.3361894786,
0.5540936589,
0.0682713017,
0.7764111757,
0.1837372631,
-0.4282325506,
0.4321610332,
-0.0065113218,
-0.5359595418,
-0.1082092896,
-0.6073560715,
-0.0429838225,
-0.1878602952,
1.2386131287,
0.0900765583,
0.0878986865,
0.2112104595,
-0.... |
http://arxiv.org/abs/2406.11651v1 | http://arxiv.org/pdf/2406.11651v1 | A Two-dimensional Zero-shot Dialogue State Tracking Evaluation Method using GPT-4 | 2024-06-17T15:32:17 | 2024-06-17T15:32:17 | Ming Gu, Yan Yang | cs.CL | cs.CL | Dialogue state tracking (DST) is evaluated by exact matching methods, which
rely on large amounts of labeled data and ignore semantic consistency, leading
to over-evaluation. Currently, leveraging large language models (LLM) in
evaluating natural language processing tasks has achieved promising results.
However, using ... | [
-0.1455634236,
-0.9227077365,
1.2123966217,
-0.0275641456,
-0.5657895207,
0.2681461871,
-0.0696535856,
-0.2425130159,
0.0319081768,
0.4015002251,
-0.7327517867,
-0.579986155,
-0.5314173102,
0.0063790842,
-0.0385986008,
1.0889055729,
-0.3144402802,
-0.0302985869,
0.0037009437,
0... |
http://arxiv.org/abs/2407.00066v1 | http://arxiv.org/pdf/2407.00066v1 | Compress then Serve: Serving Thousands of LoRA Adapters with Little Overhead | 2024-06-17T15:21:35 | 2024-06-17T15:21:35 | Rickard Brüel-Gabrielsson, Jiacheng Zhu, Onkar Bhardwaj, Leshem Choshen, Kristjan Greenewald, Mikhail Yurochkin, Justin Solomon | cs.DC | cs.DC, cs.AI, cs.CL, cs.LG | Fine-tuning large language models (LLMs) with low-rank adapters (LoRAs) has
become common practice, often yielding numerous copies of the same LLM
differing only in their LoRA updates. This paradigm presents challenges for
systems that serve real-time responses to queries that each involve a different
LoRA. Prior works... | [
-0.6028857827,
-0.5765377879,
0.3924657106,
0.2169922739,
-0.3591780365,
-0.1321959049,
-0.2624221742,
-0.5414863825,
0.6435952783,
0.7618086934,
-0.3977382779,
0.0163826384,
-0.5339189172,
-0.0327239819,
-0.1611872166,
0.5220373273,
0.2462956458,
-0.24391222,
-0.1473725587,
-0... |
http://arxiv.org/abs/2406.11634v1 | http://arxiv.org/pdf/2406.11634v1 | The Base-Rate Effect on LLM Benchmark Performance: Disambiguating Test-Taking Strategies from Benchmark Performance | 2024-06-17T15:14:10 | 2024-06-17T15:14:10 | Kyle Moore, Jesse Roberts, Thao Pham, Oseremhen Ewaleifoh, Doug Fisher | cs.CL | cs.CL, cs.AI | Cloze testing is a common method for measuring the behavior of large language
models on a number of benchmark tasks. Using the MMLU dataset, we show that the
base-rate probability (BRP) differences across answer tokens are significant
and affect task performance ie. guess A if uncertain. We find that
counterfactual pro... | [
-0.6196225286,
-0.9650720954,
0.0802708715,
0.0552128814,
-0.1345521808,
-0.1853721738,
-0.1275800914,
-0.5165287852,
-0.0119376536,
0.047147084,
-0.5957124233,
-0.0484943241,
-0.5942299962,
0.127844736,
-0.1255630702,
0.6052129865,
0.0064499076,
0.0845456347,
-0.2845385373,
-0... |
http://arxiv.org/abs/2406.11632v1 | http://arxiv.org/pdf/2406.11632v1 | Unveiling the Power of Source: Source-based Minimum Bayes Risk Decoding for Neural Machine Translation | 2024-06-17T15:13:52 | 2024-06-17T15:13:52 | Boxuan Lyu, Hidetaka Kamigaito, Kotaro Funakoshi, Manabu Okumura | cs.CL | cs.CL, cs.AI | Maximum a posteriori decoding, a commonly used method for neural machine
translation (NMT), aims to maximize the estimated posterior probability.
However, high estimated probability does not always lead to high translation
quality. Minimum Bayes Risk (MBR) decoding offers an alternative by seeking
hypotheses with the h... | [
0.0475654975,
-0.1211725101,
0.5434350371,
0.1490917802,
-0.5427630544,
0.1536628604,
0.088136822,
0.0855650753,
0.1157869026,
0.5760336518,
-0.4842779338,
-0.4250338078,
-0.8890365362,
0.0527513623,
-0.224026531,
0.9594026208,
0.2298986763,
0.2994706631,
0.0138461702,
0.044887... |
http://arxiv.org/abs/2406.11629v3 | http://arxiv.org/pdf/2406.11629v3 | Can Many-Shot In-Context Learning Help Long-Context LLM Judges? See More, Judge Better! | 2024-06-17T15:11:58 | 2024-06-30T13:31:24 | Mingyang Song, Mao Zheng, Xuan Luo | cs.CL | cs.CL | Leveraging Large Language Models (LLMs) as judges for judging the performance
of LLMs has recently garnered attention. However, this type of approach is
affected by the potential biases in LLMs, raising concerns about the
reliability of the evaluation results. To mitigate this issue, we propose and
study two versions o... | [
-0.4334945679,
-0.5147017241,
0.9025990367,
0.2708021998,
-0.4459919631,
-0.1979373842,
-0.4177382886,
-0.4866036177,
-0.2804800272,
0.2979904711,
-0.2535875738,
-0.4541969895,
-0.6544121504,
0.2194494605,
0.0508781485,
1.2470611334,
-0.1689638346,
0.4930133522,
-0.2047013491,
... |
http://arxiv.org/abs/2406.11624v1 | http://arxiv.org/pdf/2406.11624v1 | Words in Motion: Representation Engineering for Motion Forecasting | 2024-06-17T15:07:55 | 2024-06-17T15:07:55 | Omer Sahin Tas, Royden Wagner | cs.LG | cs.LG, cs.CL, cs.CV | Motion forecasting transforms sequences of past movements and environment
context into future motion. Recent methods rely on learned representations,
resulting in hidden states that are difficult to interpret. In this work, we
use natural language to quantize motion features in a human-interpretable way,
and measure th... | [
-0.1947691292,
-0.102528505,
0.7402237058,
0.2272914946,
-0.3119919598,
0.3732307553,
0.4144607782,
-0.4719523489,
0.5097542405,
0.6090308428,
-0.9274902344,
-0.1201421097,
-0.7283940911,
-0.3116663992,
-0.8249877691,
0.5824939609,
-0.3513482809,
0.1850768477,
-0.1028150618,
-0... |
http://arxiv.org/abs/2406.11622v1 | http://arxiv.org/pdf/2406.11622v1 | Building Knowledge-Guided Lexica to Model Cultural Variation | 2024-06-17T15:05:43 | 2024-06-17T15:05:43 | Shreya Havaldar, Salvatore Giorgi, Sunny Rai, Thomas Talhelm, Sharath Chandra Guntuku, Lyle Ungar | cs.CL | cs.CL | Cultural variation exists between nations (e.g., the United States vs.
China), but also within regions (e.g., California vs. Texas, Los Angeles vs.
San Francisco). Measuring this regional cultural variation can illuminate how
and why people think and behave differently. Historically, it has been
difficult to computatio... | [
-0.5290727615,
-0.6963902116,
0.4662665725,
0.283741802,
-0.0567021035,
0.4442743957,
-0.2130133808,
-0.6687021852,
0.6827261448,
0.2856693268,
-0.6853265166,
-0.5094477534,
-0.7128884792,
0.0850993246,
-0.0696589202,
0.6975137591,
-0.0832709298,
0.1973866969,
-0.1344456375,
-0... |
http://arxiv.org/abs/2406.11617v1 | http://arxiv.org/pdf/2406.11617v1 | DELLA-Merging: Reducing Interference in Model Merging through Magnitude-Based Sampling | 2024-06-17T15:02:45 | 2024-06-17T15:02:45 | Pala Tej Deep, Rishabh Bhardwaj, Soujanya Poria | cs.CL | cs.CL | With the proliferation of domain-specific models, model merging has emerged
as a set of techniques that combine the capabilities of multiple models into
one that can multitask without the cost of additional training. In this paper,
we propose a new model merging technique, Drop and rEscaLe via sampLing with
mAgnitude (... | [
-0.9630795121,
-0.4060118198,
0.1776594818,
-0.0622741766,
-0.100717023,
-0.2945750654,
-0.037167944,
-0.4046082497,
0.078352958,
0.4084367752,
-0.93583709,
-0.1776173711,
-0.338830471,
-0.0843434185,
-0.7211485505,
1.04738307,
0.2315305024,
0.1788438261,
-0.321359247,
0.130723... |
http://arxiv.org/abs/2406.11614v1 | http://arxiv.org/pdf/2406.11614v1 | Intrinsic Evaluation of Unlearning Using Parametric Knowledge Traces | 2024-06-17T15:00:35 | 2024-06-17T15:00:35 | Yihuai Hong, Lei Yu, Shauli Ravfogel, Haiqin Yang, Mor Geva | cs.CL | cs.CL, cs.AI | The task of "unlearning" certain concepts in large language models (LLMs) has
attracted immense attention recently, due to its importance for mitigating
undesirable model behaviours, such as the generation of harmful, private, or
incorrect information. Current protocols to evaluate unlearning methods largely
rely on be... | [
-0.0910927802,
-0.690094769,
0.5067806244,
-0.1336738318,
-0.375623703,
0.3568072319,
-0.4437276125,
0.0555107109,
-0.3416559994,
0.4486307204,
-0.3696798384,
-0.6842948198,
-0.4281766415,
-0.4415553212,
-0.7768285275,
0.9671630859,
-0.071594663,
0.0530712828,
-0.6884361506,
-0... |
http://arxiv.org/abs/2406.11598v1 | http://arxiv.org/pdf/2406.11598v1 | Understanding "Democratization" in NLP and ML Research | 2024-06-17T14:47:06 | 2024-06-17T14:47:06 | Arjun Subramonian, Vagrant Gautam, Dietrich Klakow, Zeerak Talat | cs.CL | cs.CL, cs.CY | Recent improvements in natural language processing (NLP) and machine learning
(ML) and increased mainstream adoption have led to researchers frequently
discussing the "democratization" of artificial intelligence. In this paper, we
seek to clarify how democratization is understood in NLP and ML publications,
through lar... | [
-0.4600512683,
-0.850840807,
0.2872650027,
-0.0227151141,
-0.3595323861,
0.1407749802,
-0.5067493916,
-0.42389974,
0.3874883354,
0.4515589178,
-0.1927797645,
-0.2246664762,
-0.5927467942,
0.2449893355,
-0.3193496764,
1.0363175869,
0.0223093163,
0.2415689379,
-0.1687109321,
0.01... |
http://arxiv.org/abs/2406.11581v1 | http://arxiv.org/pdf/2406.11581v1 | Style Transfer with Multi-iteration Preference Optimization | 2024-06-17T14:20:53 | 2024-06-17T14:20:53 | Shuai Liu, Jonathan May | cs.CL | cs.CL | Numerous recent techniques for text style transfer characterize their
approaches as variants of reinforcement learning and preference optimization.
In this work, we consider the relationship between these approaches and a class
of optimization approaches developed primarily for (non-neural) statistical
machine translat... | [
-0.3414550424,
-0.5233111978,
0.5413922668,
0.3303768933,
-0.2431985587,
0.1748520881,
-0.6515756249,
-0.1634177417,
0.384719342,
0.2905421853,
-0.7071108818,
-0.1036071405,
-0.7383217216,
0.3640197515,
-0.2580571175,
1.2762415409,
-0.3731769025,
-0.0676994026,
-0.0361668728,
-... |
http://arxiv.org/abs/2406.11580v1 | http://arxiv.org/pdf/2406.11580v1 | Error Span Annotation: A Balanced Approach for Human Evaluation of Machine Translation | 2024-06-17T14:20:47 | 2024-06-17T14:20:47 | Tom Kocmi, Vilém Zouhar, Eleftherios Avramidis, Roman Grundkiewicz, Marzena Karpinska, Maja Popović, Mrinmaya Sachan, Mariya Shmatova | cs.CL | cs.CL | High-quality Machine Translation (MT) evaluation relies heavily on human
judgments. Comprehensive error classification methods, such as Multidimensional
Quality Metrics (MQM), are expensive as they are time-consuming and can only be
done by experts, whose availability may be limited especially for low-resource
language... | [
-0.4072700441,
-0.4137945175,
0.2376358509,
0.9181700945,
-0.0883535594,
0.4570495784,
-0.0521946065,
-0.7407557964,
0.2580669522,
0.3522780836,
-0.5262190104,
-0.2622376978,
-0.815798223,
0.4753565192,
-0.3139889836,
0.654120326,
-0.5688638687,
0.0183755513,
0.0411105938,
-0.4... |
http://arxiv.org/abs/2406.11577v1 | http://arxiv.org/pdf/2406.11577v1 | Mathematical Entities: Corpora and Benchmarks | 2024-06-17T14:11:00 | 2024-06-17T14:11:00 | Jacob Collard, Valeria de Paiva, Eswaran Subrahmanian | cs.CL | cs.CL, math.HO | Mathematics is a highly specialized domain with its own unique set of
challenges. Despite this, there has been relatively little research on natural
language processing for mathematical texts, and there are few mathematical
language resources aimed at NLP. In this paper, we aim to provide annotated
corpora that can be ... | [
-0.5396437645,
-0.862547636,
0.3628872931,
0.4781813323,
-0.2107977122,
-0.0201751497,
-0.3380116522,
-0.5879837871,
0.0804819539,
0.3500729501,
-0.5232229233,
-0.5302665234,
-0.5033720732,
0.4774923921,
-0.2576631308,
0.8532853127,
-0.1673171818,
0.378892988,
-0.6538137197,
-0... |
http://arxiv.org/abs/2406.11568v1 | http://arxiv.org/pdf/2406.11568v1 | Towards an End-to-End Framework for Invasive Brain Signal Decoding with Large Language Models | 2024-06-17T14:04:18 | 2024-06-17T14:04:18 | Sheng Feng, Heyang Liu, Yu Wang, Yanfeng Wang | cs.CL | cs.CL, cs.SD, eess.AS, q-bio.NC | In this paper, we introduce a groundbreaking end-to-end (E2E) framework for
decoding invasive brain signals, marking a significant advancement in the field
of speech neuroprosthesis. Our methodology leverages the comprehensive
reasoning abilities of large language models (LLMs) to facilitate direct
decoding. By fully i... | [
-0.2326728255,
-0.977920115,
0.6159125566,
0.3085099757,
-0.241776824,
-0.1352833062,
-0.3126493692,
-0.7116631866,
0.3634477854,
0.3706111014,
-0.4676351845,
-0.5671674013,
-0.8060294986,
-0.0319950432,
-0.5298659205,
0.5773952007,
0.0573694035,
0.1552831233,
-0.0856654942,
0.... |
http://arxiv.org/abs/2406.11566v1 | http://arxiv.org/pdf/2406.11566v1 | MEMLA: Enhancing Multilingual Knowledge Editing with Neuron-Masked Low-Rank Adaptation | 2024-06-17T14:03:50 | 2024-06-17T14:03:50 | Jiakuan Xie, Pengfei Cao, Yuheng Chen, Yubo Chen, Kang Liu, Jun Zhao | cs.CL | cs.CL | Knowledge editing aims to adjust the knowledge within large language models
(LLMs) to prevent their responses from becoming obsolete or inaccurate.
However, existing works on knowledge editing are primarily conducted in a
single language, which is inadequate for multilingual language models. In this
paper, we focus on ... | [
-0.2557899952,
-0.6897227168,
0.2606095076,
0.312250942,
-0.0447542071,
0.2697972655,
-0.4950047135,
-0.43188411,
0.4228584766,
0.2662759721,
-0.3006095886,
-0.672172606,
-0.7317180037,
0.0969043225,
0.1792617887,
0.6767058969,
-0.0111471266,
0.2105629146,
-0.1821484715,
-0.497... |
http://arxiv.org/abs/2406.11565v2 | http://arxiv.org/pdf/2406.11565v2 | Extrinsic Evaluation of Cultural Competence in Large Language Models | 2024-06-17T14:03:27 | 2024-06-19T05:18:56 | Shaily Bhatt, Fernando Diaz | cs.CL | cs.CL, cs.CY | Productive interactions between diverse users and language technologies
require outputs from the latter to be culturally relevant and sensitive. Prior
works have evaluated models' knowledge of cultural norms, values, and
artifacts, without considering how this knowledge manifests in downstream
applications. In this wor... | [
-0.2968377769,
-0.581612885,
0.2712458074,
0.042359449,
-0.1797108501,
0.0406009108,
-0.7561398149,
-0.5124329925,
-0.0304351021,
0.4094307721,
-0.4837550223,
-0.2512561679,
-0.2312271148,
0.1513921022,
-0.2932722867,
1.2373911142,
-0.2990559638,
-0.3189746439,
-0.4485955238,
-... |
http://arxiv.org/abs/2406.11555v1 | http://arxiv.org/pdf/2406.11555v1 | Input Conditioned Graph Generation for Language Agents | 2024-06-17T13:53:15 | 2024-06-17T13:53:15 | Lukas Vierling, Jie Fu, Kai Chen | cs.CL | cs.CL, cs.AI | Recent progress in Large Language Models (LLMs) and language agents has
demonstrated significant promise for various future applications across
multiple disciplines. While traditional approaches to language agents often
rely on fixed, handcrafted designs, our research aims to develop both learnable
and dynamic agents. ... | [
-0.3377414644,
-1.0778068304,
0.5995106697,
0.0578847453,
-0.0833089799,
0.558447957,
-0.4715802073,
-0.2626728117,
0.211289227,
0.3686525226,
-0.7137770653,
-0.6329760551,
-0.5905055404,
0.1730796248,
-0.0322045013,
0.9587727189,
-0.1284141093,
-0.1890686154,
-0.2044687867,
-0... |
http://arxiv.org/abs/2406.11547v1 | http://arxiv.org/pdf/2406.11547v1 | GECOBench: A Gender-Controlled Text Dataset and Benchmark for Quantifying Biases in Explanations | 2024-06-17T13:44:37 | 2024-06-17T13:44:37 | Rick Wilming, Artur Dox, Hjalmar Schulz, Marta Oliveira, Benedict Clark, Stefan Haufe | cs.LG | cs.LG, cs.AI, cs.CL, cs.CY | Large pre-trained language models have become popular for many applications
and form an important backbone of many downstream tasks in natural language
processing (NLP). Applying 'explainable artificial intelligence' (XAI)
techniques to enrich such models' outputs is considered crucial for assuring
their quality and sh... | [
-0.5523385406,
-0.6331787705,
0.3187616765,
-0.0105868895,
0.0131059187,
-0.1549619585,
-0.2124829739,
-0.3818437755,
-0.3329335153,
0.296518743,
-0.8236143589,
-0.4921721518,
-0.4724543393,
-0.1023834348,
-0.6415278912,
1.1036164761,
-0.0611326881,
0.1846593767,
-0.4356866777,
... |
http://arxiv.org/abs/2406.11546v1 | http://arxiv.org/pdf/2406.11546v1 | GigaSpeech 2: An Evolving, Large-Scale and Multi-domain ASR Corpus for Low-Resource Languages with Automated Crawling, Transcription and Refinement | 2024-06-17T13:44:20 | 2024-06-17T13:44:20 | Yifan Yang, Zheshu Song, Jianheng Zhuo, Mingyu Cui, Jinpeng Li, Bo Yang, Yexing Du, Ziyang Ma, Xunying Liu, Ziyuan Wang, Ke Li, Shuai Fan, Kai Yu, Wei-Qiang Zhang, Guoguo Chen, Xie Chen | eess.AS | eess.AS, cs.CL, cs.SD | The evolution of speech technology has been spurred by the rapid increase in
dataset sizes. Traditional speech models generally depend on a large amount of
labeled training data, which is scarce for low-resource languages. This paper
presents GigaSpeech 2, a large-scale, multi-domain, multilingual speech
recognition co... | [
-0.2690244019,
-0.6898062229,
0.0245715436,
0.1526323408,
-0.2016155869,
0.4682419896,
-0.3680915236,
-0.5703746676,
0.409103781,
0.4682913423,
-0.194315806,
0.0784209818,
-0.7138504386,
-0.0928481594,
-0.4436509907,
0.8181720376,
0.1498843879,
0.1790440828,
-0.0773979425,
-0.0... |
http://arxiv.org/abs/2406.11514v1 | http://arxiv.org/pdf/2406.11514v1 | Counterfactual Debating with Preset Stances for Hallucination Elimination of LLMs | 2024-06-17T13:21:23 | 2024-06-17T13:21:23 | Yi Fang, Moxin Li, Wenjie Wang, Hui Lin, Fuli Feng | cs.CL | cs.CL | Large Language Models (LLMs) excel in various natural language processing
tasks but struggle with hallucination issues. Existing solutions have
considered utilizing LLMs' inherent reasoning abilities to alleviate
hallucination, such as self-correction and diverse sampling methods. However,
these methods often overtrust... | [
-0.9296091795,
-1.1326850653,
0.9298967719,
0.0071798046,
-0.2969979048,
-0.0641022176,
-0.3225282729,
-0.4090614617,
0.4895499647,
0.8302232027,
-0.62950629,
-0.2811619937,
-0.502707243,
0.2787111104,
-0.3215264678,
0.9316028357,
-0.1673825681,
0.1755815148,
-0.5433676243,
-0.... |
http://arxiv.org/abs/2406.11930v1 | http://arxiv.org/pdf/2406.11930v1 | A Critical Study of What Code-LLMs (Do Not) Learn | 2024-06-17T13:11:17 | 2024-06-17T13:11:17 | Abhinav Anand, Shweta Verma, Krishna Narasimhan, Mira Mezini | cs.SE | cs.SE, cs.AI, cs.CL | Large Language Models trained on code corpora (code-LLMs) have demonstrated
impressive performance in various coding assistance tasks. However, despite
their increased size and training dataset, code-LLMs still have limitations
such as suggesting codes with syntactic errors, variable misuse etc. Some
studies argue that... | [
-0.335169971,
-0.3903206587,
0.6453811526,
0.1109566614,
-0.0273584239,
0.335305512,
-0.5037770271,
-0.5940003395,
0.0913344324,
0.4470082521,
-0.4236028492,
-0.6582064629,
-0.7687453628,
0.3706440032,
-0.3382028341,
1.112459898,
-0.0433920883,
0.171125114,
-0.3694879413,
-0.11... |
http://arxiv.org/abs/2406.11503v1 | http://arxiv.org/pdf/2406.11503v1 | GeoGPT4V: Towards Geometric Multi-modal Large Language Models with Geometric Image Generation | 2024-06-17T13:04:27 | 2024-06-17T13:04:27 | Shihao Cai, Keqin Bao, Hangyu Guo, Jizhi Zhang, Jun Song, Bo Zheng | cs.CV | cs.CV, cs.CL | Large language models have seen widespread adoption in math problem-solving.
However, in geometry problems that usually require visual aids for better
understanding, even the most advanced multi-modal models currently still face
challenges in effectively using image information. High-quality data is crucial
for enhanci... | [
-0.3143827617,
-0.761783123,
0.9057804942,
0.0064995387,
-0.182836026,
-0.269993782,
-0.2213329971,
-0.2402905524,
-0.2792509198,
0.3507853448,
-0.4257518351,
-0.7628563046,
-0.4897366166,
-0.0335407481,
-0.1206712797,
0.92228055,
-0.1641528606,
0.2130566537,
-0.1385926157,
-0.... |
http://arxiv.org/abs/2406.11497v2 | http://arxiv.org/pdf/2406.11497v2 | CrAM: Credibility-Aware Attention Modification in LLMs for Combating Misinformation in RAG | 2024-06-17T13:01:12 | 2024-06-27T10:18:53 | Boyi Deng, Wenjie Wang, Fengbin Zhu, Qifan Wang, Fuli Feng | cs.CL | cs.CL | Retrieval-Augmented Generation (RAG) can alleviate hallucinations of Large
Language Models (LLMs) by referencing external documents. However, the
misinformation in external documents may mislead LLMs' generation. To address
this issue, we explore the task of "credibility-aware RAG", in which LLMs
automatically adjust t... | [
-0.2762804627,
-0.8806919456,
0.439162612,
0.3230041265,
-0.2483255714,
0.34065938,
-0.3888999224,
-0.6165131927,
0.3491581678,
0.7380799651,
-0.2235133052,
-0.3314894736,
-0.7444900274,
-0.0556659847,
-0.377504766,
1.0294904709,
0.0837560967,
-0.0173587669,
-0.433893621,
-0.10... |
http://arxiv.org/abs/2406.11486v1 | http://arxiv.org/pdf/2406.11486v1 | Analysing zero-shot temporal relation extraction on clinical notes using temporal consistency | 2024-06-17T12:53:21 | 2024-06-17T12:53:21 | Vasiliki Kougia, Anastasiia Sedova, Andreas Stephan, Klim Zaporojets, Benjamin Roth | cs.CL | cs.CL, cs.LG | This paper presents the first study for temporal relation extraction in a
zero-shot setting focusing on biomedical text. We employ two types of prompts
and five LLMs (GPT-3.5, Mixtral, Llama 2, Gemma, and PMC-LLaMA) to obtain
responses about the temporal relations between two events. Our experiments
demonstrate that LL... | [
0.3055554628,
-1.0359712839,
1.3408093452,
0.5003384352,
-0.4200738966,
-0.0028859051,
0.1102001071,
-0.8284427524,
0.3650806248,
0.1606672406,
-0.7158549428,
-0.5465666652,
-0.7093679309,
0.3090463579,
-0.1798206866,
1.1004303694,
-0.1813438237,
0.1222575754,
-0.1537903547,
-0... |
http://arxiv.org/abs/2406.11477v1 | http://arxiv.org/pdf/2406.11477v1 | Vocabulary Expansion for Low-resource Cross-lingual Transfer | 2024-06-17T12:42:34 | 2024-06-17T12:42:34 | Atsuki Yamaguchi, Aline Villavicencio, Nikolaos Aletras | cs.CL | cs.CL, cs.AI | Large language models (LLMs) have shown remarkable capabilities in many
languages beyond English. Yet, LLMs require more inference steps when
generating non-English text due to their reliance on English-centric
tokenizers, vocabulary, and pre-training data, resulting in higher usage costs
to non-English speakers. Vocab... | [
-0.2031074166,
-0.5517482162,
0.1853072345,
0.2484728098,
-0.292614013,
0.3005230725,
-0.7209798098,
-0.5221427679,
0.1408534199,
0.1474533379,
-0.207274884,
-0.2633166909,
-0.7200277448,
0.0940435454,
-0.0573201291,
0.797480762,
-0.0662193224,
0.0172170028,
-0.0864634663,
-0.1... |
http://arxiv.org/abs/2406.11474v1 | http://arxiv.org/pdf/2406.11474v1 | How Far Can In-Context Alignment Go? Exploring the State of In-Context Alignment | 2024-06-17T12:38:48 | 2024-06-17T12:38:48 | Heyan Huang, Yinghao Li, Huashan Sun, Yu Bai, Yang Gao | cs.CL | cs.CL, cs.AI | Recent studies have demonstrated that In-Context Learning (ICL), through the
use of specific demonstrations, can align Large Language Models (LLMs) with
human preferences known as In-Context Alignment (ICA), indicating that models
can comprehend human instructions without requiring parameter adjustments.
However, the e... | [
-0.3671786487,
-0.6637089849,
0.7449594736,
-0.1049135104,
-0.393155694,
-0.0155038023,
-0.2702059448,
-0.6807284951,
-0.3340452313,
0.144732058,
-0.407117188,
-0.3100544512,
-0.250598222,
0.0975318328,
-0.5429810882,
0.9697497487,
-0.2865896523,
0.0209535174,
-0.5020660162,
-0... |
http://arxiv.org/abs/2406.11473v1 | http://arxiv.org/pdf/2406.11473v1 | Promises, Outlooks and Challenges of Diffusion Language Modeling | 2024-06-17T12:38:38 | 2024-06-17T12:38:38 | Justin Deschenaux, Caglar Gulcehre | cs.CL | cs.CL, cs.AI | The modern autoregressive Large Language Models (LLMs) have achieved
outstanding performance on NLP benchmarks, and they are deployed in the real
world. However, they still suffer from limitations of the autoregressive
training paradigm. For example, autoregressive token generation is notably slow
and can be prone to \... | [
-0.225575909,
-0.8924486637,
0.7269482613,
0.3740199506,
-0.0638608634,
-0.1064719334,
-0.0908959433,
-0.21959804,
-0.0969728827,
0.2671681643,
-0.4512805641,
-0.3554141819,
-0.8633602262,
-0.0821042806,
-0.6373192072,
0.7376908064,
-0.102255024,
-0.1283573955,
-0.2043954134,
0... |
http://arxiv.org/abs/2406.11464v1 | http://arxiv.org/pdf/2406.11464v1 | Automating Easy Read Text Segmentation | 2024-06-17T12:25:25 | 2024-06-17T12:25:25 | Jesús Calleja, Thierry Etchegoyhen, David Ponce | cs.CL | cs.CL | Easy Read text is one of the main forms of access to information for people
with reading difficulties. One of the key characteristics of this type of text
is the requirement to split sentences into smaller grammatical segments, to
facilitate reading. Automated segmentation methods could foster the creation of
Easy Read... | [
-0.2616394758,
-0.9446883202,
0.8282876611,
0.2368771285,
-0.4829999804,
0.0427080803,
-0.0079039717,
-0.7507221103,
-0.0324604474,
0.5241882205,
-0.6105026007,
-0.4166692495,
-0.3047909439,
0.8750691414,
-0.4837317169,
0.5677019954,
-0.0972087979,
-0.15777421,
-0.0686659887,
-... |
http://arxiv.org/abs/2406.11460v1 | http://arxiv.org/pdf/2406.11460v1 | TRACE the Evidence: Constructing Knowledge-Grounded Reasoning Chains for Retrieval-Augmented Generation | 2024-06-17T12:23:32 | 2024-06-17T12:23:32 | Jinyuan Fang, Zaiqiao Meng, Craig Macdonald | cs.CL | cs.CL, cs.AI | Retrieval-augmented generation (RAG) offers an effective approach for
addressing question answering (QA) tasks. However, the imperfections of the
retrievers in RAG models often result in the retrieval of irrelevant
information, which could introduce noises and degrade the performance,
especially when handling multi-hop... | [
-0.3350487351,
-0.6732981801,
0.9724636674,
-0.3514387608,
-0.1848705262,
0.4814016223,
0.5672011375,
-0.4207196832,
0.3832799792,
0.5000368953,
-0.2500114143,
-0.1540820301,
-0.4766192734,
-0.166760698,
-0.4962939322,
0.9547717571,
0.023340974,
-0.1615970135,
-0.2510224283,
-0... |
http://arxiv.org/abs/2406.11455v1 | http://arxiv.org/pdf/2406.11455v1 | Adaptive Reinforcement Learning Planning: Harnessing Large Language Models for Complex Information Extraction | 2024-06-17T12:11:01 | 2024-06-17T12:11:01 | Zepeng Ding, Ruiyang Ke, Wenhao Huang, Guochao Jiang, Yanda Li, Deqing Yang, Yanghua Xiao, Jiaqing Liang | cs.CL | cs.CL, cs.AI | Existing research on large language models (LLMs) shows that they can solve
information extraction tasks through multi-step planning. However, their
extraction behavior on complex sentences and tasks is unstable, emerging issues
such as false positives and missing elements. We observe that decomposing
complex extractio... | [
-0.3360436559,
-0.9947474599,
0.6316131353,
0.4074186981,
-0.139329195,
0.2386306077,
-0.4397925436,
-0.3593798876,
-0.1738738865,
0.5743027329,
-0.5895362496,
-0.3620339334,
-0.7679358721,
0.107495755,
-0.1486984193,
0.8030699492,
-0.201289922,
0.0660538003,
0.3496298492,
0.18... |
http://arxiv.org/abs/2406.11431v1 | http://arxiv.org/pdf/2406.11431v1 | Super(ficial)-alignment: Strong Models May Deceive Weak Models in Weak-to-Strong Generalization | 2024-06-17T11:36:39 | 2024-06-17T11:36:39 | Wenkai Yang, Shiqi Shen, Guangyao Shen, Zhi Gong, Yankai Lin | cs.CL | cs.CL, cs.AI | Superalignment, where humans are weak supervisors of superhuman models, has
become an important and widely discussed issue in the current era of rapid
development of Large Language Models (LLMs). The recent work preliminarily
studies this problem by using weak models to supervise strong models. It
discovers that weakly... | [
-0.1685256511,
-0.5567597747,
0.5626088381,
-0.2095549554,
-0.0681880414,
0.0427406207,
0.08939749,
-0.5875198841,
0.0973967314,
0.3672705889,
-0.6127628684,
-0.0667272657,
-0.7315508723,
-0.2491014749,
-0.4830227792,
1.1786948442,
0.0830885544,
0.3213231862,
-0.3072544038,
-0.... |
http://arxiv.org/abs/2406.11430v1 | http://arxiv.org/pdf/2406.11430v1 | A Simple and Effective $L_2$ Norm-Based Strategy for KV Cache Compression | 2024-06-17T11:35:16 | 2024-06-17T11:35:16 | Alessio Devoto, Yu Zhao, Simone Scardapane, Pasquale Minervini | cs.CL | cs.CL, cs.AI | The deployment of large language models (LLMs) is often hindered by the
extensive memory requirements of the Key-Value (KV) cache, especially as
context lengths increase. Existing approaches to reduce the KV cache size
involve either fine-tuning the model to learn a compression strategy or
leveraging attention scores t... | [
-0.2999854088,
-0.33488065,
0.4761122167,
0.2925587893,
-0.5643543005,
-0.2589292526,
-0.5191634893,
-0.2969220877,
0.0478479229,
0.3468830884,
-0.3339496851,
-0.4186713398,
-0.5902537704,
-0.2494220138,
-0.1812486649,
0.4336492419,
0.2259036303,
-0.1145666391,
-0.3285272121,
-... |
http://arxiv.org/abs/2406.11429v1 | http://arxiv.org/pdf/2406.11429v1 | Fusion Makes Perfection: An Efficient Multi-Grained Matching Approach for Zero-Shot Relation Extraction | 2024-06-17T11:31:48 | 2024-06-17T11:31:48 | Shilong Li, Ge Bai, Zhang Zhang, Ying Liu, Chenji Lu, Daichi Guo, Ruifang Liu, Yong Sun | cs.CL | cs.CL, cs.AI | Predicting unseen relations that cannot be observed during the training phase
is a challenging task in relation extraction. Previous works have made progress
by matching the semantics between input instances and label descriptions.
However, fine-grained matching often requires laborious manual annotation, and
rich inte... | [
-0.4828926921,
-0.7297818661,
0.8186739683,
0.0598873757,
-0.111728102,
-0.2240861505,
0.2116674781,
-0.949072957,
0.1717515439,
0.5958604217,
-0.4951059222,
-0.261064738,
-0.3230672777,
-0.0877820253,
-0.4048099816,
0.9361911416,
0.0303479582,
-0.1726830751,
-0.5158692002,
-0.... |
http://arxiv.org/abs/2406.11427v1 | http://arxiv.org/pdf/2406.11427v1 | DiTTo-TTS: Efficient and Scalable Zero-Shot Text-to-Speech with Diffusion Transformer | 2024-06-17T11:25:57 | 2024-06-17T11:25:57 | Keon Lee, Dong Won Kim, Jaehyeon Kim, Jaewoong Cho | eess.AS | eess.AS, cs.AI, cs.CL, cs.LG, cs.SD | Large-scale diffusion models have shown outstanding generative abilities
across multiple modalities including images, videos, and audio. However,
text-to-speech (TTS) systems typically involve domain-specific modeling factors
(e.g., phonemes and phoneme-level durations) to ensure precise temporal
alignments between tex... | [
-0.2393502593,
-0.8938125968,
0.6404997706,
0.1344396472,
-0.5048782825,
0.1653131992,
-0.0377781838,
0.0559263118,
0.3055133522,
-0.0471840724,
-0.4650459886,
-0.2792056799,
-1.1049081087,
0.052298829,
-0.8904564381,
0.9050649405,
0.211904183,
-0.0776196048,
0.1667806953,
0.11... |
http://arxiv.org/abs/2406.11424v1 | http://arxiv.org/pdf/2406.11424v1 | Evaluating the Efficacy of Open-Source LLMs in Enterprise-Specific RAG Systems: A Comparative Study of Performance and Scalability | 2024-06-17T11:22:25 | 2024-06-17T11:22:25 | Gautam B, Anupam Purwar | cs.IR | cs.IR, cs.CL | This paper presents an analysis of open-source large language models (LLMs)
and their application in Retrieval-Augmented Generation (RAG) tasks, specific
for enterprise-specific data sets scraped from their websites. With the
increasing reliance on LLMs in natural language processing, it is crucial to
evaluate their pe... | [
-0.3777153194,
-0.8025490642,
0.6513960361,
0.3224267066,
-0.0599937811,
0.0970916301,
-0.405113101,
-0.6957304478,
0.405403614,
0.7197762728,
-0.0391803011,
-0.5043756366,
-0.2625706792,
0.2124225199,
-0.4784934223,
0.6262708306,
-0.0036255247,
0.184269771,
-0.3335846961,
-0.3... |
http://arxiv.org/abs/2406.11423v1 | http://arxiv.org/pdf/2406.11423v1 | Dredge Word, Social Media, and Webgraph Networks for Unreliable Website Classification and Identification | 2024-06-17T11:22:04 | 2024-06-17T11:22:04 | Evan M. Williams, Peter Carragher, Kathleen M. Carley | cs.SI | cs.SI, cs.AI, cs.CL, cs.CY, cs.LG | In an attempt to mimic the complex paths through which unreliable content
spreads between search engines and social media, we explore the impact of
incorporating both webgraph and large-scale social media contexts into website
credibility classification and discovery systems. We further explore the usage
of what we def... | [
-0.1641031206,
-1.3201146126,
0.4464049041,
0.2036241293,
-0.2873100936,
0.7271360159,
0.0213905219,
-0.6370175481,
0.4912280738,
0.0593529008,
-0.5084827542,
-0.3524248004,
-0.8893897533,
-0.1987059712,
-0.6184718013,
0.7227891088,
-0.0895106345,
-0.0425983779,
-0.0880464837,
... |
http://arxiv.org/abs/2406.11418v1 | http://arxiv.org/pdf/2406.11418v1 | BAMBINO-LM: (Bilingual-)Human-Inspired Continual Pretraining of BabyLM | 2024-06-17T11:08:08 | 2024-06-17T11:08:08 | Zhewen Shen, Aditya Joshi, Ruey-Cheng Chen | cs.CL | cs.CL | Children from bilingual backgrounds benefit from interactions with parents
and teachers to re-acquire their heritage language. In this paper, we
investigate how this insight from behavioral study can be incorporated into the
learning of small-scale language models. We introduce BAMBINO-LM, a continual
pretraining strat... | [
-0.2162090242,
-0.7568712831,
-0.0058476627,
0.0996571779,
-0.4579376876,
0.3965837657,
-0.1864264607,
-0.4719753861,
0.1200968847,
0.3097792566,
-0.580286324,
-0.0958262905,
-0.5211409926,
0.0280309971,
0.098910749,
1.2290837765,
0.0866039172,
0.3071939945,
0.4029137492,
0.078... |
http://arxiv.org/abs/2406.15486v2 | http://arxiv.org/pdf/2406.15486v2 | SampleAttention: Near-Lossless Acceleration of Long Context LLM Inference with Adaptive Structured Sparse Attention | 2024-06-17T11:05:15 | 2024-06-28T08:55:17 | Qianchao Zhu, Jiangfei Duan, Chang Chen, Siran Liu, Xiuhong Li, Guanyu Feng, Xin Lv, Huanqi Cao, Xiao Chuanfu, Xingcheng Zhang, Dahua Lin, Chao Yang | cs.CL | cs.CL, cs.AI, cs.LG | Large language models (LLMs) now support extremely long context windows, but
the quadratic complexity of vanilla attention results in significantly long
Time-to-First-Token (TTFT) latency. Existing approaches to address this
complexity require additional pretraining or finetuning, and often sacrifice
model accuracy. In... | [
-0.0625510514,
-0.6973700523,
0.653595686,
0.4622285366,
-0.2596859038,
0.2126564234,
-0.3857572675,
-0.5223087072,
0.2084517926,
0.335577637,
-0.803876996,
-0.3686660826,
-0.3289674222,
-0.0548922084,
-0.4509217441,
0.8121398687,
-0.0019778092,
-0.3176536858,
-0.2253673971,
-0... |
http://arxiv.org/abs/2406.15485v2 | http://arxiv.org/pdf/2406.15485v2 | SegHist: A General Segmentation-based Framework for Chinese Historical Document Text Line Detection | 2024-06-17T11:00:04 | 2024-06-25T07:37:22 | Xingjian Hu, Baole Wei, Liangcai Gao | cs.CL | cs.CL, cs.CV | Text line detection is a key task in historical document analysis facing many
challenges of arbitrary-shaped text lines, dense texts, and text lines with
high aspect ratios, etc. In this paper, we propose a general framework for
historical document text detection (SegHist), enabling existing
segmentation-based text det... | [
-0.3641759455,
-1.0428197384,
0.5051228404,
0.0738660023,
-0.6080548167,
-0.4408370256,
0.2055089623,
-0.4684963524,
0.4831526279,
0.7651533484,
-0.7609000802,
-0.6635122299,
0.1139775068,
0.134304598,
-0.4064905047,
1.0100598335,
-0.087421082,
0.045901686,
0.123042278,
-0.2258... |
http://arxiv.org/abs/2406.11410v2 | http://arxiv.org/pdf/2406.11410v2 | HARE: HumAn pRiors, a key to small language model Efficiency | 2024-06-17T10:56:03 | 2024-06-18T11:59:03 | Lingyun Zhang, Bin jin, Gaojian Ge, Lunhui Liu, Xuewen Shen, Mingyong Wu, Houqian Zhang, Yongneng Jiang, Shiqi Chen, Shi Pu | cs.CL | cs.CL, cs.AI | Human priors play a crucial role in efficiently utilizing data in deep
learning. However, with the development of large language models (LLMs), there
is an increasing emphasis on scaling both model size and data volume, which
often diminishes the importance of human priors in data construction.
Influenced by these tren... | [
-0.3895145655,
-0.261826098,
0.581032455,
0.0434864461,
-0.1051052287,
-0.1808000654,
-0.5719495416,
-0.1630218327,
-0.1505675465,
0.7599827647,
-0.6193315983,
-0.5565809011,
-0.4052472413,
0.0062564961,
-0.0864369795,
0.672059238,
0.1729308814,
0.1404989958,
-0.2220348716,
-0.... |
http://arxiv.org/abs/2406.11409v2 | http://arxiv.org/pdf/2406.11409v2 | CodeGemma: Open Code Models Based on Gemma | 2024-06-17T10:54:35 | 2024-06-19T02:37:50 | CodeGemma Team, Heri Zhao, Jeffrey Hui, Joshua Howland, Nam Nguyen, Siqi Zuo, Andrea Hu, Christopher A. Choquette-Choo, Jingyue Shen, Joe Kelley, Kshitij Bansal, Luke Vilnis, Mateo Wirth, Paul Michel, Peter Choy, Pratik Joshi, Ravin Kumar, Sarmad Hashmi, Shubham Agrawal, Zhitao Gong, Jane Fine, Tris Warkentin, Ale Jak... | cs.CL | cs.CL, cs.AI | This paper introduces CodeGemma, a collection of specialized open code models
built on top of Gemma, capable of a variety of code and natural language
generation tasks. We release three model variants. CodeGemma 7B pretrained (PT)
and instruction-tuned (IT) variants have remarkably resilient natural language
understand... | [
-0.2621902823,
-0.7796074152,
0.125722453,
0.2589769661,
-0.1207758337,
0.1451675445,
-0.0184201784,
-0.6472434998,
-0.0431002676,
0.8111636043,
-0.4082774222,
-0.4622402191,
-0.5780015588,
-0.0441153497,
-0.1753126532,
0.8406169415,
-0.0314788148,
0.3694941998,
-0.1184848025,
... |
http://arxiv.org/abs/2406.11403v1 | http://arxiv.org/pdf/2406.11403v1 | Multimodal Structured Generation: CVPR's 2nd MMFM Challenge Technical Report | 2024-06-17T10:45:47 | 2024-06-17T10:45:47 | Franz Louis Cesista | cs.CV | cs.CV, cs.CL | Multimodal Foundation Models (MMFMs) have shown remarkable performance on
various computer vision and natural language processing tasks. However, their
performance on particular tasks such as document understanding is still
limited. They also require more compute, time, and engineering resources to
finetune and deploy ... | [
-0.2875674665,
-0.6642866135,
0.502068758,
0.5679712892,
-0.1878162026,
0.0439340733,
-0.1088487953,
-0.4452781379,
-0.3352380693,
0.6187465191,
-0.8913857341,
-0.2440723479,
-0.5553573966,
0.1918512732,
-0.3480587602,
0.7483828068,
-0.2125046253,
-0.1605410874,
-0.5958526731,
... |
http://arxiv.org/abs/2406.11402v1 | http://arxiv.org/pdf/2406.11402v1 | Evaluating Open Language Models Across Task Types, Application Domains, and Reasoning Types: An In-Depth Experimental Analysis | 2024-06-17T10:45:36 | 2024-06-17T10:45:36 | Neelabh Sinha, Vinija Jain, Aman Chadha | cs.CL | cs.CL, cs.AI, cs.LG | The rapid rise of Language Models (LMs) has expanded their use in several
applications. Yet, due to constraints of model size, associated cost, or
proprietary restrictions, utilizing state-of-the-art (SOTA) LLMs is not always
feasible. With open, smaller LMs emerging, more applications can leverage their
capabilities, ... | [
-0.3359995186,
-0.9298601747,
0.8371314406,
0.2446705103,
0.0900508836,
-0.1272850782,
-0.5508303642,
-0.4108676016,
-0.302408278,
0.5286473632,
-0.5135082006,
-0.4267272353,
-0.4628139436,
0.3239447773,
-0.280857265,
0.7424131036,
-0.3255318105,
0.0920166969,
-0.2973071337,
-0... |
http://arxiv.org/abs/2406.11400v1 | http://arxiv.org/pdf/2406.11400v1 | Large Language Models and Knowledge Graphs for Astronomical Entity Disambiguation | 2024-06-17T10:38:03 | 2024-06-17T10:38:03 | Golnaz Shapurian | cs.CL | cs.CL, astro-ph.IM | This paper presents an experiment conducted during a hackathon, focusing on
using large language models (LLMs) and knowledge graph clustering to extract
entities and relationships from astronomical text. The study demonstrates an
approach to disambiguate entities that can appear in various contexts within
the astronomi... | [
-0.4941679239,
-0.9939715862,
0.9747510552,
-0.038558498,
-0.1034264266,
0.2909644842,
-0.3732157946,
-0.8507663608,
0.5850892663,
0.5449876189,
-0.4064941406,
-0.5293229818,
-0.6632636189,
0.2435121983,
-0.7049206495,
0.7288595438,
0.0673739985,
0.0518813357,
-0.3028632402,
0.... |
http://arxiv.org/abs/2406.11385v2 | http://arxiv.org/pdf/2406.11385v2 | MetaGPT: Merging Large Language Models Using Model Exclusive Task Arithmetic | 2024-06-17T10:12:45 | 2024-06-27T16:01:28 | Yuyan Zhou, Liang Song, Bingning Wang, Weipeng Chen | cs.CL | cs.CL | The advent of large language models (LLMs) like GPT-4 has catalyzed the
exploration of multi-task learning (MTL), in which a single model demonstrates
proficiency across diverse tasks. Task arithmetic has emerged as a
cost-effective approach for MTL. It enables performance enhancement across
multiple tasks by adding th... | [
-0.4233603179,
-0.7466591001,
0.5198654532,
0.1210557148,
-0.2336613983,
-0.0407345891,
-0.2030740082,
-0.2350788414,
0.1235526055,
0.5189220905,
-0.5715069175,
-0.4286080599,
-0.7998579144,
-0.1547599882,
-0.2943928242,
1.077914834,
-0.2641661763,
0.0243369602,
-0.1851503104,
... |
http://arxiv.org/abs/2406.11380v1 | http://arxiv.org/pdf/2406.11380v1 | A Realistic Evaluation of LLMs for Quotation Attribution in Literary Texts: A Case Study of LLaMa3 | 2024-06-17T09:56:46 | 2024-06-17T09:56:46 | Gaspard Michel, Elena V. Epure, Romain Hennequin, Christophe Cerisara | cs.CL | cs.CL | Large Language Models (LLMs) zero-shot and few-shot performance are subject
to memorization and data contamination, complicating the assessment of their
validity. In literary tasks, the performance of LLMs is often correlated to the
degree of book memorization. In this work, we carry out a realistic evaluation
of LLMs ... | [
0.3161111772,
-0.6065596342,
0.7357573509,
0.195651263,
-0.4357189238,
-0.0372563004,
-0.0826365799,
-0.4251646399,
0.0773991495,
0.726549983,
-0.3180386424,
-0.4534293413,
-0.3498649895,
0.2280374914,
-0.5147622824,
0.9863772988,
-0.3402824402,
-0.2860901356,
-0.5623339415,
0.... |
http://arxiv.org/abs/2406.11375v1 | http://arxiv.org/pdf/2406.11375v1 | Boosting Scientific Concepts Understanding: Can Analogy from Teacher Models Empower Student Models? | 2024-06-17T09:51:38 | 2024-06-17T09:51:38 | Siyu Yuan, Cheng Jiayang, Lin Qiu, Deqing Yang | cs.CL | cs.CL, cs.AI | Analogical reasoning plays a critical role in human cognition, enabling us to
understand new concepts by associating them with familiar ones. Previous
research in the AI community has mainly focused on identifying and generating
analogies and then examining their quality under human evaluation, which
overlooks the prac... | [
-0.1332406551,
-0.4702912271,
0.8407065868,
-0.1583049744,
-0.3647483587,
0.224188298,
-0.0199651774,
-0.5649657249,
0.2825648487,
0.2081820816,
-0.0229107775,
0.0905913338,
-0.1716414839,
0.2680171728,
-0.418732971,
0.6650267243,
0.0130686928,
0.5329704285,
-0.9336262345,
0.04... |
http://arxiv.org/abs/2406.11370v1 | http://arxiv.org/pdf/2406.11370v1 | Fairer Preferences Elicit Improved Human-Aligned Large Language Model Judgments | 2024-06-17T09:48:53 | 2024-06-17T09:48:53 | Han Zhou, Xingchen Wan, Yinhong Liu, Nigel Collier, Ivan Vulić, Anna Korhonen | cs.CL | cs.CL, cs.AI, cs.CY, cs.LG | Large language models (LLMs) have shown promising abilities as cost-effective
and reference-free evaluators for assessing language generation quality. In
particular, pairwise LLM evaluators, which compare two generated texts and
determine the preferred one, have been employed in a wide range of
applications. However, L... | [
-0.2180299163,
-0.3722297549,
0.7904200554,
0.0413498729,
-0.1725285798,
-0.2081760019,
-0.452577889,
-0.3551403284,
-0.139137581,
0.3079318702,
-0.358907342,
-0.3642595708,
-0.4745280445,
0.2146821469,
0.2715183794,
0.8800243735,
0.3282963932,
0.1490300745,
-0.206515789,
-0.07... |
http://arxiv.org/abs/2406.11368v1 | http://arxiv.org/pdf/2406.11368v1 | Improving Quotation Attribution with Fictional Character Embeddings | 2024-06-17T09:46:35 | 2024-06-17T09:46:35 | Gaspard Michel, Elena V. Epure, Romain Hennequin, Christophe Cerisara | cs.CL | cs.CL | Humans naturally attribute utterances of direct speech to their speaker in
literary works. When attributing quotes, we process contextual information but
also access mental representations of characters that we build and revise
throughout the narrative. Recent methods to automatically attribute such
utterances have exp... | [
-0.0064671258,
-0.2699562311,
0.4937215447,
-0.0441385619,
-0.281858176,
0.0863357857,
-0.1919971257,
-0.58940804,
0.2310193926,
0.7171458602,
-0.4174774587,
-0.2521808147,
-0.1993855536,
0.2066488117,
-0.0345088392,
0.842944324,
0.1134339795,
0.0776882246,
-0.5655412078,
-0.15... |
http://arxiv.org/abs/2406.16937v1 | http://arxiv.org/pdf/2406.16937v1 | A Complete Survey on LLM-based AI Chatbots | 2024-06-17T09:39:34 | 2024-06-17T09:39:34 | Sumit Kumar Dam, Choong Seon Hong, Yu Qiao, Chaoning Zhang | cs.CL | cs.CL, cs.AI | The past few decades have witnessed an upsurge in data, forming the
foundation for data-hungry, learning-based AI technology. Conversational
agents, often referred to as AI chatbots, rely heavily on such data to train
large language models (LLMs) and generate new content (knowledge) in response
to user prompts. With th... | [
-0.1753767133,
-0.940328002,
0.4141190052,
-0.0856213719,
0.3761297464,
0.3745245039,
-0.1882995218,
-0.4975502789,
0.1072712615,
0.7165611386,
-0.5456522703,
-0.3591910899,
-0.5601229668,
-0.2139276862,
-0.4633349478,
0.8343150616,
0.2696006894,
-0.1808709651,
-0.6627143621,
-... |
http://arxiv.org/abs/2406.11357v2 | http://arxiv.org/pdf/2406.11357v2 | Refiner: Restructure Retrieval Content Efficiently to Advance Question-Answering Capabilities | 2024-06-17T09:25:10 | 2024-06-18T02:44:27 | Zhonghao Li, Xuming Hu, Aiwei Liu, Kening Zheng, Sirui Huang, Hui Xiong | cs.CL | cs.CL, cs.AI | Large Language Models (LLMs) are limited by their parametric knowledge,
leading to hallucinations in knowledge-extensive tasks. To address this,
Retrieval-Augmented Generation (RAG) incorporates external document chunks to
expand LLM knowledge. Furthermore, compressing information from document chunks
through extractio... | [
-0.4701681435,
-0.5624716282,
0.6312999725,
-0.2306368798,
-0.2557623982,
0.3254016042,
-0.3261211812,
-0.2792277038,
0.4398730099,
0.6638667583,
-0.522126615,
-0.0251041185,
-0.3636477888,
0.2088047415,
-0.6127242446,
1.0130938292,
-0.0149004795,
-0.2091554105,
-0.4199654162,
... |
http://arxiv.org/abs/2406.12937v1 | http://arxiv.org/pdf/2406.12937v1 | Self-Train Before You Transcribe | 2024-06-17T09:21:00 | 2024-06-17T09:21:00 | Robert Flynn, Anton Ragni | eess.AS | eess.AS, cs.CL, cs.LG, cs.SD | When there is a mismatch between the training and test domains, current
speech recognition systems show significant performance degradation.
Self-training methods, such as noisy student teacher training, can help address
this and enable the adaptation of models under such domain shifts. However,
self-training typically... | [
-0.1663073599,
-0.2181674987,
0.0152003542,
0.0608216897,
-0.1241909191,
0.2752830982,
-0.2111667544,
-0.3964638114,
-0.2763815224,
0.3844079077,
-0.54006809,
0.0422842577,
-0.5637415051,
-0.030517891,
-0.6663116813,
0.8043296933,
0.0483174808,
0.373023659,
-0.0822514221,
-0.09... |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.