Hypothesis Evolution Protocol proposed for auditable AI scientist agents
A new arXiv paper introduces the Hypothesis Evolution Protocol (HEP), a framework to make LLM-based AI scientists auditable by structuring hypothesis generation, testing, and belief updates in a transparent, logged process. The protocol aims to address the lack of auditability in current autonomous scientific agents.
Entities
Related
arXiv preprint introduces IdeaGene-Bench for scientific lineage reasoning
A new benchmark, IdeaGene-Bench (IG-Bench), evaluates AI systems on scientific lineage reasoning and idea generation grounded in prior work. It frames scientific ideas as inheriting mechanisms and recombining earlier pieces, akin to biological genomes.
Researchers propose semantic framework to classify AI system failures
A new arXiv paper introduces a semantic framework for describing AI systems, distinguishing justified outputs from common failures like extrapolation, refuted assertions, and stale sources. The framework aims to help practitioners systematically evaluate correctness of AI-generated representations.
GRACE: Graph-Regularized Agentic Context Evolution for Reliable Long-Horizon LLM Agents
A new arXiv paper proposes GRACE, a method that maintains persistent system-level instructions for LLM agents as a typed semantic graph instead of flat text. This graph-regularized approach enables scoped verification and reliable context evolution over long horizons under distribution shift, addressing verification difficulties from accumulated instructions.
Yohei Nakajima publishes paper 'The Log is the Agent' proposing log-driven agent design
Yohei Nakajima released a paper titled 'The Log is the Agent' that rethinks agent architecture by treating the log as the core of the agent rather than a debugging afterthought. The approach inverts the standard build order of chat loop, tool calling, rules, and logging, suggesting the log itself should drive agent behavior.
ProofCouncil: An LLM Agent for Solving Open Mathematical Problems
Researchers introduced ProofCouncil, an LLM agent with an author-critic architecture designed to solve open mathematical problems. It was submitted to the FirstProof challenge, where it autonomously tackled 6 out of 10 problems and received referee evaluations.