Search the full wire by company, model, lab, or keyword. Every story we have ever aggregated.
ProcArena benchmark evaluates LLMs on multi-scenario PL/SQL development including direct generation, modification, debugging, and interactive multi-turn workflows.
Study characterizes interaction between inference-time activation steering and weight-only quantization (INT8, NF4) on 7-9B open-weight LLMs.
Cross-step control framework mitigates higher-order sequential interference in multi-domain RL training for LLMs by analyzing consecutive gradient reversals.
Activation steering in LLMs validates whether steering vectors encode coherent human-value geometry using Schwartz's moral theory.
Audit of 22 frontier LLMs: widespread digit-level retrieval of published values on molecular benchmarks, conflating memorization with prediction.
GUT method quantifies reasoning uncertainty in LLMs by modeling branching complexity as graph structure, enabling detection of nonsensical reasoning chains.
Knowledge Space Theory framework evaluates whether LLMs exhibit structured, prerequisite-dependent knowledge in mathematical reasoning.
Mixture-of-Experts (MoE) architectures provide an efficient paradigm for scaling large language models (LLMs), yet fixed top-k routing activates the same number of expert slots for every token, causing substantial redundant computation. Existing expert-skipping methods often…
On-policy distillation of LLMs recovers full-data performance gains with single-query training via state coverage and alignment dynamics.
LLMs over-edit code during repair; study of 400 BigCodeBench problems shows widespread over-editing even in frontier models.
FLY-EVAL++: safety-focused evaluation protocol for LLMs in physics-governed domains; measures constraint violations and physical inconsistency beyond accuracy metrics.
InSituMeasure benchmark evaluates MLLMs on continuous-valued measurement tasks in realistic industrial settings with gauge reading and instrument-specific context.
Diffusion-augmented LLMs enable parallel token generation from autoregressive distributions, decoupling NTP and diffusion weights for lossless speedup.
Demonstrates user feedback signals strong LLM improvement signal; prior evaluation methods systematically underestimated utility via synthetic and naturalistic data.
"The United States has a strong interest in continuing to develop a robust and competitive artificial intelligence industry that sets the standard for the practice and procedure of AI use globally," the brief reads.
Structured reasoning framework for LLM-powered telecom root cause analysis with evidence grounding to reduce hallucination.
ShallowStream optimizes streaming video MLLMs by pruning computation at shallow layers, reducing overhead for embodied AI and autonomous driving.
MedMisBench study reveals how LLMs fail under misleading medical context via fabricated evidence; implications for clinical AI safety.
DKL combines finetuning with retrieval-augmented generation to inject corpus-specific knowledge into instruction-tuned LLMs without requiring massive synthetic data.
Hybrid hallucination detection for black-box LLMs combines semantic entropy and token log-probability uncertainty to reduce false positives and missed fabrications.
LLMs automate vulnerability injection into Solidity smart contracts to generate ground-truth datasets for testing vulnerability detection tools across 49 OpenSCV types.
Replicates seminal economic experiments with LLM agents in double auction markets to test whether human-designed mechanisms deliver efficient resource allocation.
Answer-verified multi-teacher distillation identifies correct teacher per sample rather than per domain, improving multi-domain LLM integration from domain-matched supervision.
Persona prompting steers LLMs toward survey-participant response alignment; empirical study of attribute effectiveness for population prediction.
Debias-SparseGPT addresses bias amplification in pruned LLMs using representational debiasing with demographically contrasting inputs.
Comparative analysis of worldbuilding strategies in AI-generated vs. human fiction across GPT-4, LLaMA 3.3, Mistral 3.2, Gemma 3.
Behavior-based fusion model combining LLM differential diagnoses with ontology rankers for rare-disease diagnosis interpretability.
Retrieval-augmented generation and BioNER improve lay summarization of radiology reports to reduce patient reliance on general LLMs.
Causal analysis across 9 open-weight models shows quantization damage is distributed globally, not localized to task circuits or weight statistics.
Framework characterizing near-optimal SFT-RL annotation budget allocation across LLM sizes, showing allocation ratios don't transfer uniformly between model scales.