arXiv Computation and Language

Listening to the Wise Few: Query-Key Alignment Unlocks Latent Correct Answers in Large Language Models

arXiv Computation and Language

Sep 15

Identifying Crucial Attention Heads for Multilingual Language Models: Retrieval and Retrieval-Transition Heads

arXiv:2602.22453v4 Announce Type: replace Abstract: Retrieval heads, a subset of attention heads in Transformers, were studied in English, showing its crucial role in retrieving information from the...

By Shaswat Patel, Vishvesh Trivedi, Yue Han, Yihuai Hong, Eunsol Choi

llms benchmarks

arXiv Computation and Language

1d ago

RoPE at the End of Its Rope? Theory, Diagnosis, and Mitigation of Long-Context Failures

arXiv:2609.39929v1 Announce Type: cross Abstract: Long-context failures of RoPE-based language models can arise from RoPE's intrinsic tradeoff between maintaining stable token preferences and disting...

By Yuyang Wu, Yufeng Du, Hao Peng

llms benchmarks

arXiv AI

Aug 18

Measuring Reward Hacking and Reasoning-Answer Decoupling Under Position-Confounded Optimization

arXiv:2608. 15445v1 Announce Type: new Abstract: When a reward is correct on every training example yet consistent with more than one goal, a model can acquire an unintended one, a failure known as goal misgeneralization.

By Suyash Maniyar, Armaan Sandhu, Abhishek Mishra

llms benchmarks

arXiv AI

Jul 2

Logit-Contribution Scoring Identifies Non-Literal Retrieval Heads

arXiv:2607. 01002v1 Announce Type: cross Abstract: In long-context use, large language models frequently synthesize answers from the meaning of a relevant context span rather than literally copy-pasting them.

By Aryo Pradipta Gema, Beatrice Alex, Pasquale Minervini

llms benchmarks

arXiv AI

Aug 7

Beyond Sequence Order: Syntax-Informed Positional Embeddings for Transformers

arXiv:2608. 06111v1 Announce Type: cross Abstract: Positional embeddings (PE) in Transformers encode token distance and order but are largely agnostic to \textit{syntactic structure}.

By Haris Riaz, Hyungji Kim, Mihai Surdeanu

llms rag benchmarks

arXiv Machine Learning

Jun 2

Test-Time Compute for Frozen Embedding Models through Agentic Program Search

arXiv:2605. 11374v5 Announce Type: replace Abstract: Test-time compute is widely believed to benefit only large reasoning models, leaving small models with nothing to gain.

By Han Xiao

llms rag agents

arXiv AI

Jul 20

RLearner-LLM: Balancing Logical Grounding and Fluency in Large Language Models via Hybrid Direct Preference Optimization

arXiv:2605. 04539v4 Announce Type: replace-cross Abstract: Direct Preference Optimization (DPO), the efficient alternative to PPO-based RLHF, falls short on knowledge-intensive generation: standard preference signals from human annotators or LLM judges exhibit a systematic verbosity bias that rewards fluency over logical correctness.

By Qiming Bao, Juho Leinonen, Paul Denny, Michael J. Witbrock

llms reinforcement-learning fine-tuning safety

arXiv Machine Learning

Sep 22

Guiding the coarse levels of semantic IDs makes the fine levels learnable

arXiv:2609.22227v1 Announce Type: cross Abstract: Generative retrieval represents each item by a short Semantic ID and casts recommendation as autoregressive generation of that sequence. Because the...

By Bin Wang, Zhengyu Zhang

llms rag nlp reinforcement-learning safety

arXiv AI

Jul 7

Detecting Hallucinations in Retrieval-Augmented Generation through Grounding-Aware Sensitivity by Perturbation (GASP)

arXiv:2607. 04223v1 Announce Type: cross Abstract: Retrieval-augmented generation (RAG) reduces but does not eliminate hallucination, and existing detectors return a single answer-level score that does not indicate which sentence is unsupported, or why.

By Mohamed Aly Bouke

rag nlp benchmarks safety

arXiv AI

4d ago

Stepwise Intrinsic Rewards for Reasoning in Large Language Models

The paper introduces Stepwise Marginal Information Gain (MIG), an intrinsic process reward that evaluates how each reasoning step of a large language model (LLM) or vision-language model (VLM) improves the likelihood of the reference answer. MIG rewards only new likelihood maxima, preventing duplicate credit, and is combined with outcome, format, and self‑distillation objectives to guide training. Experiments on eight benchmarks show that this method outperforms outcome‑only reinforcement learning and improves accuracy by up to 4.8 points over binary‑reward training, including a 12.6‑point gain on MathVerse and a 12.9‑point advantage on vision‑language transfer at 7B parameters.

By Xiangwei Wang, Wei Wang, Ken Chen, Nanduni Nimalsiri, Sachith Seneviratne, Saman Halgamuge

llms rag reinforcement-learning efficiency multimodal benchmarks

arXiv AI

Jun 24

SURGELLM: Rethinking Multi-Task Evaluation through Task-Aware Feature Gating with Class-Balanced Normalization

arXiv:2606. 24259v1 Announce Type: cross Abstract: Fine-tuned encoders deployed across heterogeneous NLP tasks face three compounding problems: mismatched inductive biases, class-imbalance corruption of feature statistics, and no mechanism to condition attention on external lexical knowledge.

By Noor Islam S. Mohammad, Ulug Bayazit

llms nlp fine-tuning efficiency safety

arXiv AI

Jul 17

Introspection Fine-Tuning (IFT): Training Small LLMs to Introspect

arXiv:2607. 14111v1 Announce Type: cross Abstract: Can small language models detect and report on perturbations their own internal activations?

By Ely Hahami, Ishaan Sinha, Lavik Jain

llms fine-tuning benchmarks safety

Related stories