Community feed
A focused stream of recent stories from the sources curated for this community. Latest: Learning Topology-Aware Representations for Mechanics-Relevant TPMS Metamaterials, Causal Syntax Circuits in CodeLMs: Budget and Probe Decide What Looks Localised, and TurboMamba: Faster Inference, Faster Training, and Better Prediction with Lexical Memory. Page 31.
Planet AI Developer Filter
Abstract page for arXiv paper 2610.20684: Learning Topology-Aware Representations for Mechanics-Relevant TPMS Metamaterials
Planet AI Developer Filter
Abstract page for arXiv paper 2610.20683: Causal Syntax Circuits in CodeLMs: Budget and Probe Decide What Looks Localised
Planet AI Developer Filter
Abstract page for arXiv paper 2610.20682: TurboMamba: Faster Inference, Faster Training, and Better Prediction with Lexical Memory
Planet AI Developer Filter
Abstract page for arXiv paper 2610.20639: Same Answer, New Box: How Much of RLVR's Out-of-Domain Gain Is Answer Format?
Planet AI Developer Filter
Abstract page for arXiv paper 2610.20638: Feature Elicitation for SAE Steering of Deceptive Behaviour in Small LLMs
Planet AI Developer Filter
Abstract page for arXiv paper 2610.20780: ARIA: Runtime Inspection and Amendment for Multi-Agent Systems
Planet AI Developer Filter
Abstract page for arXiv paper 2610.20764: When Reality Outpaces Cognition: A New Scaling Principle for Reliable Intelligence
Planet AI Developer Filter
Abstract page for arXiv paper 2610.20685: L2G-SR: Learning to Guide LLM-based Symbolic Regression
Planet AI Developer Filter
Abstract page for arXiv paper 2610.20681: IneqAgent: Agentic Inequality Proving in Lean with Symbolic Computation
Planet AI Developer Filter
Abstract page for arXiv paper 2610.20679: Counterfactual Neighborhood Risk Fields: Evidence-Grounded Selective Safety for Vision-Language Models
Planet AI Developer Filter
Abstract page for arXiv paper 2610.20676: When Correctness Feedback Erases Evidence: Phase-Separated Completeness Auditing in Verus
Planet AI Developer Filter
Abstract page for arXiv paper 2610.20675: AlphaDiana: A Measurement System for Agents in Verifiable Reasoning Tasks