Search the full wire by company, model, lab, or keyword. Every story we have ever aggregated.
Study of derived-feature over-trust (DFOT) in LLMs using physiological sensing; proposes privileged-modality reliability evidence to mitigate misapplication.
PACE hierarchical framework applies LLMs to parent-order execution in algorithmic trading, generalizing across market conditions without task-specific training.
Fairness Pruning localizes demographic bias in GLU-MLP layers via differential neuron activations, enabling targeted bias mitigation in LLMs.
PathView-Bench evaluates multimodal LLMs on fine-grained, multiscale pathology image understanding with spatial annotations across 23 datasets.
HARGO framework applies heterogeneity-aware RL post-training to optimize LLM behavior on HPC tasks like data-race detection beyond SFT accuracy.
CACHE-UK memory editor for sequential fact updates in 4-bit quantized LLMs deployed in finance, addressing quantization stability crisis.
Empirical evaluation of 6 leading LLMs on Islamic knowledge tasks reveals hallucination and authenticity gaps in religious contexts.
Vibe-FDTR agent framework enables LLMs to automate reproducible thermal property analysis from natural language requests.
Compressed LLMs pass quality guards but hallucinate procedure steps in agentic execution, exposing fidelity-safety gap.
ParliamentBench evaluates deceptive reasoning in 16 LLMs via Secret Hitler game framework with 1,600 adversarial matches.
Comparative study finds LLMs and linguist trainees struggle similarly with subjective Appraisal theory annotation in TED talk discourse.
It is impossible to make large language models fully secure against hacks because of a fundamental flaw in how they work, a team of researchers argue in a paper presented at the International Conference on Machine Learning, a top AI conference, this month. The claim has huge…
MMAC: 5,638-clip benchmark for audio captioning across 15 evaluation dimensions, targeting fine-grained free-form descriptions for AudioLLMs.
Stereotypes-to-Decisions framework measures regional bias in six LLMs across China's 34 provinces, linking abstract stereotypes to concrete allocation decisions.
Latent-IM recovers state estimation and action control for conversational moves in speech LLMs by decomposing move selection and realization in hidden representations.
Preregistered experiment: diverse human groups (L1/L2 English writers) produce more collective creativity than LLMs; AI assistance risks homogenizing or replacing human diversity.
Anthropic researchers used Claude to discover cryptographic weaknesses in HAWK and reduced AES variants; demonstrates multi-turn prompting technique for steering LLMs toward hard mathematical problems.
Study finds instruction-tuned LLMs exhibit stronger syntactic convergence to human dialogue patterns than humans themselves across diverse grammatical constructions.
Polistemics: theory-grounded benchmark for evaluating LLMs as political information mediators via epistemic modesty standard.
ClinMM-Bench evaluates multimodal LLMs on multi-turn clinical diagnosis with progressive information disclosure, mirroring real-world diagnostic reasoning.
Input-only prompt optimization suppresses evaluation-awareness latents in LLMs via GCG-style token search, threatening safety evaluation validity if undetected.
Empirical analysis of attention patterns in LLM-based automated program repair to explain inconsistent patch generation performance.
Evaluation of adversarial robustness in five Arabic language models under multi-granularity attacks exposes security vulnerabilities in non-English LLMs.
OmniDelta allocates token budgets by skill across modalities in OmniLLMs, compressing audio-video sequences for inference efficiency.
Construction-Driven Injection embeds linguistically-grounded code-mixing fingerprints in LLMs to create verifiable ownership signals resistant to accidental activation and perplexity filtering.
BeyondUncertainty uses LLM confidence estimates to route retrieval for knowledge-intensive QA, reducing irrelevant evidence and computation.
MMOE applies sparse-expert efficiency principles from LLMs to diffusion transformers, balancing generation quality against training and deployment cost.
LLM-SoccerArena: live prospective benchmark measuring LLM forecasting on real-world sports outcomes pre-event.
Adapts MLLMs (Qwen3-VL) to sparse 8-16 frame video input for temporal grounding, addressing training-deployment mismatch.
Framework for auditing multimodal LLM faithfulness in power grid diagnosis via modality ablation and behavioral reliance testing.