Large language model (LLM) news — GPT-4, Claude, Gemini, Llama, Mistral and the latest research on training, fine-tuning, RLHF, and deployment of LLMs.
Researchers introduced SLARouter, an online routing algorithm for large language model (LLM) applications that learns cost-optimal policies from sparse user feedback while guaranteeing Service Level Agreement (SLA) compl…
Researchers propose AgenticRei, a deontic policy framework for runtime governance of LLM-driven autonomous AI agents, addressing obligations, dispensations, conflict resolution, and ontological reasoning beyond current p…
A systematic experimental analysis of eight state-of-the-art Diffusion Language Models (DLMs) across eight benchmarks reveals that generation-time design choices strongly influence performance and computational efficienc…
Researchers model multi-agent LLM deliberation as a closed-loop dynamical system where each agent has a hidden internal belief, or anchor, that pulls its opinion. They show the anchor can be recovered from deliberation a…
Researchers found that large language models (LLMs) fail to recognize their own knowledge limits on clinical tabular data, outputting near-constant confidence scores regardless of accuracy. By analyzing attribution diver…
A new study measures the linearity of transformer feed-forward blocks, finding that linear recoverability (R²_lin) varies widely across blocks and is a learned property, not an architectural one. The analysis of GPT-2, P…
Researchers introduced a prompt-based uncertainty decomposition method that separates action confidence from request uncertainty, enabling LLM agents to proactively seek clarification when task specifications are ambiguo…
Researchers identified a 'narration gap' in LLM-solver loops where formal solver outputs are narrated by language models, potentially compromising soundness. Prompt injection attacks can invert verified conclusions acros…
Researchers at arXiv propose a framework for selecting the most informative comparison pairs in preference-based post-training of large language models, showing that strategic pair selection can improve sample efficiency…
Researchers argue that aggregate-score leaderboards for LLM agent benchmarks systematically underspecify deployed-agent evaluation, as rankings do not transfer to out-of-distribution settings. They propose ranking config…
Researchers introduced GLARE, an LLM-based natural language interface that translates user questions into SQL queries over local explanation data to provide global explanations for black-box image classifiers. The system…
Researchers introduced GRACE, a theoretical framework that determines the optimal verification granularity for test-time scaling in large language models based on problem difficulty, verifier accuracy, and compute budget…
Researchers introduced TreeTracer, a visual analytics tool that aggregates hundreds of stochastic LLM generations into syntax-aligned trees to expose hidden biases. The tool uses perturbation analysis and contrastive inf…
Researchers developed an ensemble of large language models, including Google's Gemini and Gemma, to automatically identify EQ-5D health-related quality-of-life studies in PubMed from abstracts. The weighted ensemble achi…
A study fine-tuning seven large language models on Arabic found no evidence of Semitic-specific cross-lingual transfer; gains were driven by task-format alignment rather than linguistic relatedness.
A new study introduces an error taxonomy for LLMs in hardware design, revealing that frontier models plateau at a 90.8% pass rate on the VerilogEval benchmark due to unsolvable functional errors. The research shows that …
DeepSeek AI released preview versions of its DeepSeek-V4 series, including two Mixture-of-Experts language models with up to 1.6 trillion parameters and support for one-million-token contexts. The models feature architec…
Researchers have identified and characterized positional bias in in-context learning for diffusion large language models (dLLMs), showing that query placement significantly affects generation quality. They propose Averag…
Researchers introduced the Argent Signaling Protocol (ASP) to mitigate semantic drift in multi-agent LLM systems by attaching structured quality signals to AI responses. In tests, ASP improved pass rates on a document-gr…
Researchers at the Allen Institute for AI used training-data attribution to map which regions of the pretraining corpus support social-reasoning versus STEM-reasoning in the OLMo3-7B language model. They found that socia…