Search the full wire by company, model, lab, or keyword. Every story we have ever aggregated.
Large language models (LLMs) are known to abandon their initial stance to conform to user pushback. While prior research largely attributes this behavior to sycophancy learned during reinforcement learning from human feedback, we hypothesize that conformity is also driven by a…
Modern Large Language Models (LLMs) are often criticized for producing repetitive and homogeneous text, despite possessing vast latent vocabularies. While previous research has focused on model knowledge and training data, we investigate the role of decoding mechanics in…
Long chain-of-thought reasoning has made autoregressive decoding the dominant inference cost of modern large language models. Existing methods target either the input side (latent compression) or the output side (speculative decoding and multi-token prediction, MTP), but the…
An effective method of teaching across disciplines is to provide examples of high-quality work. However, an example may be significantly different from a student's current work, making it challenging for them to emulate. An ideal learning demonstration is a counterfactual…
Recent advances in large language models (LLMs) have facilitated the widespread deployment of LLMs as interactive agents capable of reasoning, planning, and tool use. Despite strong performance on existing benchmarks, such agents often exhibit notable degradation when deployed…
Animation elevates digital documents into immersive experiences, yet creating custom motion paths remains cumbersome, requiring designers to manually select presets, plot Bézier points, and configure timing properties. We introduce Generative Animations, a system that…
Large language models often solve tasks from a fully specified prompt but degrade when the same requirements unfold over multiple turns, known as the lost-in-conversation (LiC) gap. We trace part of this degradation to self-contamination: intermediate assistant replies enter…
Method conditions diffusion models on MLLMs for subject-driven image generation, improving identity preservation and instruction following via joint text-image encoding.
LLM-based framework for structured code change labeling (renames, moves, logic edits) to improve code review efficiency beyond summarization.
Auto Benchmark Audit uses agentic framework to systematically audit AI benchmarks for hidden dependencies, specification gaps, and grading flaws.
>Long-context inference in large language models is bottlenecked by the quadratic cost of full attention. Existing efficient alternatives often rely either on native sparse training or on heuristic token eviction, creating an undesirable trade-off among efficiency, training…
ChartFI benchmark evaluates faithfulness and insightfulness of MLLM-generated chart descriptions beyond fact enumeration.
Corpus-linguistic evaluation framework measures linguistic humanlikeness of LLM outputs via register-aware patterns; addresses underexplored text quality dimension.
Production systems generate millions of log lines daily, yet most anomaly detectors operate at the session or window-level, flagging groups of lines rather than identifying the specific message responsible. This coarse granularity forces operators to inspect many routine lines…
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…
LASH composes complementary jailbreak attack families adaptively per-prompt to overcome single-strategy limitations.
Spanish psychiatric ICD classification dataset (145k entries) benchmarked classical NLP vs. LLMs for automated diagnosis coding.
Context-invariant safety alignment framework enforces LLM refusal behavior independent of prompt surface form, using verifiable and noisy feedback selectively.
KoRe: integrating compact knowledge graphs with LLMs to address hallucination and knowledge update via human-readable representations.
MixRea benchmark tests whether LLMs exhibit inattentional blindness to subtle contextual cues across 2,246 multi-choice questions.
Survey of 120+ studies on mathematical reasoning in LLMs: datasets, architectures, training strategies, evaluation protocols for AI benchmarking.
LLM-based generative error correction for low-resource West Frisian ASR with data contamination analysis and offline dataset construction.
Vision Inference Former addresses visual consistency in MLLMs by elevating visual features above text-token parity in connector-based architectures.
ArXiv is doing more to crack down on the careless use of large language models in scientific papers.
DebiasRAG mitigates social bias and hallucinations in LLMs through retrieval-augmented generation without fine-tuning.
LLM-driven data augmentation using GPT-5 to generate synthetic speech for cognitive score prediction from Japanese narratives.
Study finds LLMs systematically adapt linguistic behavior when perceiving social observation, with governance implications.
SemaTune uses LLM guidance for online OS tuning with semantic awareness of system knob policies.