Search the full wire by company, model, lab, or keyword. Every story we have ever aggregated.
Large language models (LLMs) exhibit systematic political bias across a variety of sensitive contexts. We find that LLMs handle counterpart topics from opposing political sides asymmetrically. We refer to this phenomenon as covert political bias and identify 7 categories of…
Large language models (LLMs) are typically trained on shuffled corpora, yielding models whose knowledge is frozen at train time and whose temporal grounding remains poorly understood. In this work, we study the impact of pre-training dynamics on the acquisition of…
Large language models (LLMs) increasingly excel at mathematical reasoning, but their unreliability limits their utility in mathematics research. A mitigation is using LLMs to generate formal proofs in languages like Lean. We perform the first large-scale evaluation of this…
Study of 75,898 API calls shows LLMs exhibit accumulated message effect bias when evaluating sequential items in single conversations.
Inverse scaling detected: more capable LLMs forecast worse on superlinear/regime-change time series; ForecastBench-Sim benchmark released.
Image-semantic detection method enhances MLLM performance detecting AI-generated modern Chinese poetry.
Interactive guide matching 50+ LLMs to 60+ hardware builds with throughput metrics and YouTube reviewer citations.
Reddit post comparing Descartes' philosophical argument against machine cognition to modern LLM capabilities; no new technical claims.
DeepWeb-Bench introduces harder evaluation for frontier LLMs on deep research requiring massive cross-source evidence and reasoning.
RLVR training exhibits rank-1 weight trajectory structure; minimal training captures performance gains via linear parameter evolution.
LLM-based grammar adaptation for metamodel evolution in domain-specific languages; evaluated on Xtext DSLs.
Reddit discussion on professional AI-assisted coding practices and code quality concerns when senior engineers use LLMs without planning or testing.
Milgram obedience variant on 11 open-source LLMs shows most models comply with authority pressure in sustained decision-making; safety concern for agents.
Question-based benchmark evaluates LLM semantic reasoning on long sequences of unstructured social media data.
Spanish psychiatric ICD classification dataset (145k entries) benchmarked classical NLP vs. LLMs for automated diagnosis coding.
APM benchmark for evaluating style personalization in LLMs using arbitrary preference mappings without reference responses.
Theoretical framework for training multimodal LLMs using only pairwise modality alignments instead of full joint multimodal datasets.
Utilizing LLMs for automated taxonomy construction presents a clear opportunity for the comprehensive, yet efficient mapping of potentially complex domains. When contending with high volumes of rapidly growing corpora, however, it becomes unclear how to best leverage such data…
System extends Text-to-SQL LLMs with agentic capability for governed enterprise APIs, handling complex business logic, auditability, and non-technical user access to analytics.
KoRe: integrating compact knowledge graphs with LLMs to address hallucination and knowledge update via human-readable representations.
Open-source tool for generating 3D objects with articulated, functional parts using LLMs, addressing limitations of diffusion-based text-to-3D pipelines.
MixRea benchmark tests whether LLMs exhibit inattentional blindness to subtle contextual cues across 2,246 multi-choice questions.
CopT reverses chain-of-thought order to draft answers before thinking, reducing token costs when LLMs solve problems without extended reasoning.
Empirical study showing that higher-fidelity observations (RGB-D vs symbolic) degrade LLM performance on embodied reasoning tasks like mechanical puzzles.
LLM method for generating multimodal agent behaviors (verbal, vocal, gestural, facial) calibrated to trustworthiness dimensions.
Framework for synthesizing long-term medical dialogues with LLMs to enable evaluation of healthcare agents reasoning over patient history.
Survey of 120+ studies on mathematical reasoning in LLMs: datasets, architectures, training strategies, evaluation protocols for AI benchmarking.
OScaR: KV cache quantization technique addressing token norm imbalance for extreme compression in long-context LLMs.
Empirical study of 2-6 bit quantization impact on eight LLMs across MMLU-Pro and reasoning benchmarks.
LLMEval-Logic: solver-verified Chinese logical reasoning benchmark with adversarial hardening and expert audit for LLM evaluation.