explore | Grilled Cheese

A new arXiv study explores whether general-purpose LLMs, orchestrated as agents, can build constraint programming models for production scheduling directly from natural-language descriptions, bridging automated modeling and agentic AI.

#AI #LLM #AIResearch
https://arxiv.org/abs/2610.10184

Bolzano, a multi-agent open-source system combining parallel prover agents with a verifier, solved about 200 open problems out of roughly 3,800 across four paper sets, including four questions from STOC 2026 papers…

#AIResearch #MathAI #OpenSource #TheoreticalCS
https://arxiv.org/abs/2610.09769

New arXiv work formalizes how to build stable behavioral variables for studying them across tasks and language agent models.

#AIResearch #LanguageAgents #Arxiv
https://arxiv.org/abs/2610.09237

New arXiv paper CADForge proposes an agentic framework that turns single-view images into editable parametric CAD models via CadQuery code, using component decomposition and explicit geometric…

#GenerativeAI #AIResearch #PromptEngineering #ComputerAidedDesign
https://arxiv.org/abs/2610.04262

FreDF is a new method that improves time series forecasting by addressing label autocorrelation through frequency-domain learning rather than the direct forecast paradigm, reducing estimation bias and outperforming existing…

#AIresearch #timeseries #machinelearning
https://arxiv.org/abs/2402.02399

A study finds that neural power-flow solvers achieve accurate voltage predictions but often produce physically inconsistent solutions, with a proposed subspace projection method helping reduce power-balance residuals.

#AI #AIResearch #Hardware
https://arxiv.org/abs/2610.05959

CHARTER is a new reference-aware evaluation charter for auditing compact-evidence explanations in computational pathology, requiring researchers to declare the target, quantify prediction shift, and audit…

#ComputationalPathology #AIResearch #MachineLearning
https://arxiv.org/abs/2610.07843

New arXiv work proposes a dual in-vehicle intrusion detection system pairing supervised and unsupervised models, with a quantised LSTM IDS hitting over 99.9% accuracy on DoS and fuzzy CAN frame attacks while staying…

#AIResearch #CyberSecurity #AutomotiveAI #IDS
https://arxiv.org/abs/2610.07489

Researchers introduced Trajectools, a no-code plugin built on the open-source MovingPandas Python library, aimed at simplifying movement data analytics within the QGIS geographic information system.

#AI #AIResearch #OpenSource
https://arxiv.org/abs/2610.06858

New arXiv paper explores how learning style shapes unlearning, finding generalization-heavy models suffer greater retain damage than memorization-heavy ones under controlled modular addition experiments.

#MachineUnlearning #AIResearch #DeepLearning
https://arxiv.org/abs/2610.08577

A dataset of 100 research-level math questions was compiled in Leipzig and evaluated across multiple LLM stages, with unsolved questions dropping from 41 to 2 after heavy-thinking model runs. arXiv:2606.05818v2

#AI #LLM #AIResearch
https://arxiv.org/abs/2606.05818

Big news! 🎉 CS faculty receive a $1 million NSF award to develop AI that uncovers the scientific laws governing complex systems.

This also marks the first federal grant for UCSB's Real AI initiative! 🏆

#UCSB #ComputerScience #AIResearch #NSF #WomenInSTEM

GEM introduces a concept erasure framework for Rectified Flow models by linking trajectory-based unlearning with teacher-guided flow matching. It targets harmful content, deepfakes, and copyright issues in…

#AIResearch #GenerativeAI #MachineLearning #AIethics
https://arxiv.org/abs/2606.00140

LLMs can match behavior without recovering the underlying mechanism—right answers, wrong reasons. New work probes this with #LLMs #AIResearch #MachineLearning #Evaluation

https://freegardner.com/synapse/behavior-matching-without-mechanism-recovery-in-llms.html

Deep neural networks across diverse tasks converge to shared low-dimensional weight subspaces, with spectral analysis of 1200+ models showing majority variance in just a few principal directions.

#AIresearch #DeepLearning #ModelWeights
https://arxiv.org/abs/2512.05117