AI/ML arXiv cs.AI

From Atomic Actions to Standard Operating Procedures: Iterative Tool Optimization for Self-Evolving LLM Agents

EvoSOP is a framework that allows LLM agents to self-evolve by synthesizing atomic actions into reusable Standard Operating Procedures (SOPs) to reduce reasoning overhead.

AI/ML arXiv cs.AI

Physics-Audited Agentic Discovery in Scientific Machine Learning

PA-SciML introduces a verification-first workflow for agentic scientific machine learning to ensure discovered surrogate models satisfy fundamental physics requirements.

AI/ML arXiv cs.AI

MIRA-Math: A Benchmark for Minimal Information Requesting and Mathematical Reasoning

MIRA-Math is a benchmark designed to evaluate an AI's ability to identify and request specifically missing atomic facts needed to solve mathematical problems.

AI/ML arXiv cs.AI

Agentic Data Environments

A proposal for 'Agentic Data Environments' reframes data systems from passive stores into active substrates that can amplify agent capabilities while enforcing safety guarantees.

AI/ML arXiv cs.AI

Reason Less, Verify More: Deterministic Gates Recover a Silent Policy-Violation Failure Mode in Tool-Using LLM Agents

Researchers found that tool-using agents often suffer from 'silent policy-violation failures' and propose using deterministic pre-execution gates to prevent these errors.

Other Hacker News

Ergo: Long Form Philosophy Lectures

A Hacker News thread discussing long-form philosophy lectures.

AI/ML arXiv cs.AI

AgentLens: Production-Assessed Trajectory Reviews for Coding Agent Evaluation

Introduction of AgentLens, an open-source benchmark for evaluating the full trajectory of interactive coding agents beyond simple pass/fail metrics.

AI/ML arXiv cs.AI

When Does In-Context Search Help? A Sampling-Complexity Theory of Reflection-Driven Reasoning

Theoretical analysis of in-context search in LLMs, demonstrating that reflection-driven reasoning can provide exponential improvements in success probability when mistakes are localized.

AI/ML arXiv cs.AI

LLM-powered reasoning in agent-based modeling

Introduction of HALE, a framework that combines Large Language Models with Agent-based Modeling to better predict human decision-making in epidemic simulations.

Hardware/Chips arXiv cs.AI

QANTIS: Hardware-Calibrated Sequential POMDP Belief Updates on IBM Heron

A hardware case study using IBM Heron quantum processors as a belief-update service for sequential POMDPs, focusing on stability and calibration.

AI/ML arXiv cs.AI

Cost-Effective Agent Harnesses for Abstract Reasoning and Generalization on ARC-AGI-1

A study on cost-effective agent harnesses for the ARC-AGI-1 benchmark, achieving high pass rates using an open-weight model without benchmark-specific fine-tuning.

AI/ML arXiv cs.AI

Evaluating SageMath-Augmented LLM Agents for Computational and Experimental Mathematics

Research evaluating the integration of SageMath into LLM agent workflows to improve performance in computational and experimental mathematics.

AI/ML arXiv cs.AI

The Harness Effect: How Orchestration Design Sets the Token Economics of Enterprise Agentic AI

Analysis of how orchestration design ('harnesses') can significantly reduce token consumption and cost in enterprise agentic AI without sacrificing quality.

AI/ML arXiv cs.AI

Grounding Spatial Relations in a Compact World Model: Instruction Leakage and a Goal-Free Dynamics Fix

Identification of 'instruction leakage' in goal-conditioned world models, where models transcribe instructions rather than perceiving relations, and a proposed fix to decouple goals from dynamics.

AI/ML arXiv cs.AI

Large Behavior Model: A Promptable Digital Twin of the Retail Customer

Presentation of a Large Behavior Model (LBM) that learns customer decision-making from retail transaction data to create promptable digital twins of customers.

AI/ML Hacker News

Benchmarking coding agents on Databricks' multi-million line codebase

Databricks provides a benchmark analysis of how coding agents perform when operating on a massive, multi-million line corporate codebase.

Software Engineering Hacker News

Cargo-nextest: 3x faster than cargo test, per-test isolation, first-class CI

Cargo-nextest is presented as a high-performance Rust test runner that offers significantly faster execution and better isolation than the standard cargo test.

AI/ML arXiv cs.AI

EVA-Net: Subject-Independent EEG Motor Decoding with Video-Derived Motor Priors

EVA-Net is a new framework for EEG motor decoding that uses action videos as semantic priors to improve cross-subject generalization in BCIs.

AI/ML arXiv cs.AI

What Type of Inference is Active Inference?

This paper provides a formal variational characterization of Expected Free Energy (EFE) based planning in active inference, clarifying the role of epistemic and planning corrections.

AI/ML arXiv cs.AI

Reducing the Complexity of Deep Learning Models for EEG Analysis on Wearable Devices

Research on deploying deep learning models for EEG analysis on wearable devices, focusing on the trade-offs between parameter quantization and accuracy for seizure detection.