OpenAI released 722 manuscripts from an unreleased frontier model covering hundreds of open math problems, with proofs formalized in Lean. Verification and credit questions now land on human referees.

#AiResearch #Lean #Mathematics #MillenniumProblems #OpenAI

A new arXiv method, FC-SWE, uses failure-conditioned trajectories in reinforcement learning to help long-horizon software engineering agents recover from failed patches by reusing diagnostic feedback from…

#SWEAgents #ReinforcementLearning #DevTools #AIResearch
https://arxiv.org/abs/2610.07898

New research examines security, privacy, and dual-use risks of Jev, an AI system for routing requests and tool selection via typed outputs and probabilities, using both cloud API and controllable local variants.

#AI #LLM #AIResearch
https://arxiv.org/abs/2610.04985

New arXiv paper studies demographic personas induced from anonymized public behavioral data for LLM-based survey simulation, examining how source domain, scale, and granularity affect alignment with real responses.

#AIResearch #LLMs #SurveySimulation
https://arxiv.org/abs/2610.05828

A new survey reviews the shift from brute-force LLM scaling toward compute-optimal training, covering parameter efficiency, quantization, LoRA, and edge-oriented methods for resource-constrained settings. The synthesis highlights techniques…

#AI #LLM #AIResearch
https://arxiv.org/abs/2610.06387

Jump Trading is using OpenAI to scale quantitative research by combining multiple data sources with human review in longer-running AI workflows. This shows how established trading firms are integrating LLMs into research pipelines…

#AI #AIResearch #OpenSource
https://openai.com/index/jump-trading

New arXiv paper introduces CCQ, a dataset of 59,372 child care provider records across 12 U.S. states, designed to support AI research on early childhood health with privacy safeguards. The work combines an…

#AIResearch #Dataset #ChildHealth #MachineLearning
https://arxiv.org/abs/2610.05863

#AlSafety #AlAlignment #TechSky
#ComputerScience #OpenSourceAl
#OpenWeight #AlgorithmicGovernance
#XRisk #AIResearch #AIGovernance
#MachineLearning #AlEthics
#DigitalProduction #Cybernetics
#Superalignment #FrontierRisk #TechTheory
#DistributedSystems

A useful AI research design for the Digital Humanities:
AI does not just make existing research faster. It shifts the boundary of what can become a feasible research question.
#DigitalHumanities #AIResearch #ComputationalHumanities

A 217 year old Napoleonic cipher has been cracked with GPT-6 Astra.
A striking example of how AI can turn previously uneconomical research problems into feasible #DigitalHumanities projects.

#DigitalHumanities #AIResearch #ComputationalHistory

arXiv paper tests whether written text alone can predict prosodic style, using acoustic clusters across six speakers and twelve text embedding models as a falsifiable hypothesis.

#AIresearch #TextToSpeech #SpeechSynthesis
https://arxiv.org/abs/2610.05575

Discovery doesn't happen in silos, and that's why MATRIX-Uni helps students explore interdisciplinary #AIresearch.

#ILLINOIS transfer student Mariia Martova worked with #NSFMMLI researchers to apply her #computerscience skills to #biochemistry.

Watch her story: youtu.be/tuSwA_b5RgU

A new pre-deployment framework estimates task difficulty for federated perception systems by combining intrinsic data properties with client-distribution composition. Experiments on three MNIST variants show strong negative…

#EdgeAI #FederatedLearning #AIResearch
https://arxiv.org/abs/2603.28282

New arXiv work proposes measuring safety-critical content via Jacobian-space readouts and introduces transport-amplification profiles linking latent safety geometry to accessible output, pointing to new training…

#cybersecurity #AIsafety #AIresearch #MLsafety
https://arxiv.org/abs/2610.04316

AIProver is an agentic framework that combines a 119B open-weight language model with an evolved tool-calling harness for proof auto-formalization and synthesis, using verifiers to address missing concepts and…

#AIResearch #AutoFormalization #Lean #MathAI
https://arxiv.org/abs/2610.05367

Deep RL just shattered its depth limit: 1000-layer networks now deliver 2-50x gains in self-supervised locomotion & manipulation, with #DeepRL #ReinforcementLearning #AIResearch #MachineLearning

https://freegardner.com/synapse/deep-rl-depth-scaling-breakthrough.html

How is #AI transforming PSP research? 🧠 Professor Larry Golbe shares how artificial intelligence is already making a difference. Read more: ow.ly/n2xA50ZTp8v

Are you using AI in your research? #PSP

#AIResearch #PSPResearch #Neuroscience #BrainResearch #NeuroTech

A new paper proposes GFGE, a framework aimed at unifying varied explainable AI methods under a common interpretive schema. It distinguishes five operational roles to clarify how evidence becomes explanation for recipients.

#XAI #MachineLearning #AIResearch
https://arxiv.org/abs/2610.05225

TasteVal is a new arXiv benchmark that scores frontier AI models on experimental research taste, defined as compute efficiency in iteratively designing experiments and drawing conclusions on fixed problems. The 8 open-ended tasks…

#AIResearch #Benchmark #AI #LLMs
https://arxiv.org/abs/2610.06824

A new paper introduces MechHypoBench, the first benchmark for evaluating AI agents' ability to generate mechanistic hypotheses from empirical data across 14 scientific fields and 17.98 million records. The authors also propose an…

#AI #AIResearch #TechNews
https://arxiv.org/abs/2610.05197