OpenAI cuts ties with three safety researchers over data policy violations. #OpenAI #AISafety #DataSecurity #AIResearch #ModelGovernance #TechEthics thedailytechfeed.com/openai-cuts-...

FORALL-LEAN-AGENT is a frontend-agnostic framework for auditable reasoning in Lean, combining isolated workspaces, axiom audits, and independent proof checking to make acceptance of candidate proofs traceable. On the…

#Lean #FormalVerification #AIResearch #DevTools
https://arxiv.org/abs/2610.00885

A new benchmark called VisionQ evaluates vision-language models on criterion-conditioned visual discrimination, using 1,800+ figures from peer-reviewed CVPR and ICCV papers to ground judgments in…

#VisionLanguageModels #ComputerVision #Benchmark #AIResearch
https://arxiv.org/abs/2610.00666

New arXiv work explores Runtime Agent Coordination, letting AI scientists pick agents and adjust division of labor during execution rather than relying on fixed workflows. Evaluated across Agent Laboratory,…

#AIResearch #MultiAgent #AIScientists #arXiv
https://arxiv.org/abs/2610.00980

Verifier leakage in agent debuggers can make solver comparisons look stronger than they are, as shown by identical hitting-set and greedy results in 12/12 dev cases until audit. Useful reminder to audit…

#AgentDebugging #SoftwareEngineering #AIResearch #DevTools
https://arxiv.org/abs/2610.00126

New arXiv paper introduces Enterprise Representation Simplification (ERS), a method for reducing representational complexity in enterprise AI systems. It pairs ERS with a model called Enterprise Representation…

#EnterpriseAI #DataManagement #AIResearch #TechNews
https://arxiv.org/abs/2610.00791

A new arXiv survey maps recent RAG research into a four-axis taxonomy covering efficiency, robustness and security, interactivity, and reasoning, moving beyond standard pipeline overviews. It highlights how retrieval-augmented generation…

#AI #LLM #AIResearch
https://arxiv.org/abs/2610.01936

New arXiv benchmark OpenMTB-Audit finds that across eight LLM configurations tested on 500 synthetic NSCLC cases, all models over-refused by dropping the Partially Supported label in 83.3–100% of true Partially Supported cases,…

#AI #LLMs #AIHealth #AIResearch
https://arxiv.org/abs/2610.01497

New arXiv paper studies cooperation in frontier generative AI using the iterated prisoner's dilemma, finding that reputation, strategy, and emotional signaling shape behavior in both reasoning and non-reasoning models. The…

#AIResearch #GenAI #Cooperation #ArXiv
https://arxiv.org/abs/2610.01222

Small language models like Qwen2.5-7B can be fine-tuned on triage data for privacy-preserving clinical decision support, with concise clinical vignettes outperforming other prompting approaches.

#AI #AIResearch #OpenSource
https://arxiv.org/abs/2604.26766

An arXiv paper surveys 5,285 NeurIPS 2025 papers and finds environmental impact reporting is nearly non-existent, proposing standardised metrics and a tool called carbonbenchmark to track LLM training and inference emissions. The authors also…

#AI #LLM #AIResearch
https://arxiv.org/abs/2610.01116

How to Understand an Unfamiliar Codebase With AI
#AITools #AIResearch #NoCode

https://askmioai.com/understand-codebase-with-ai/

A new empirical study analyzes 22,848 closed issues from 21 open-source LLM-based multi-agent systems, filtering down to 944 MAS-related issues to map practitioner challenges, their causes, and potential…

#OpenSourceAI #LLMAgents #DeveloperTools #AIResearch
https://arxiv.org/abs/2610.00905

A new benchmark called Legal Research Bench evaluates 13 frontier models on 413 expert-written US legal research questions, using agents equipped with web and case-law search tools to measure end-to-end reliability. The work highlights how a…

#AI #LLM #AIResearch
https://arxiv.org/abs/2610.00609

Generally, human-AI collaboration outperforms humans alone or AI alone. That take is not new. However, there has been surprisingly little empirical evidence across a broad set of studies to support it. Until now.

Details coming soon to scottgraffius.com.

#AI #HumanAI #HumanAITeams #AIResearch

New arXiv study examines on-device AI for real-time live-stream chat translation, highlighting CPU and thermal constraints across five mobile devices and introducing LiveChatBench, a 1,000-pair Korean-English benchmark for domain adaptation.

#AI #LLM #AIResearch
https://arxiv.org/abs/2601.02641