Self-Evolving Agent Harness defines a framework for agents that iteratively rewrite their software scaffolding - prompts, memory, workflows, tools, and executable components - while keeping core model weights fixed. It separates what can be updated (interpreted state, tools/code, repository control, plugin composition) from who proposes and who admits changes, and it implements a feedback-driven pipeline where accepted revisions persist across runs. The design treats search over harness revisions as an evolutionary optimization problem and documents practical cases of harness co-design, distributed execution, and plugin reuse, arguing that reliable deployment depends on continual validation as task distributions, evaluators, privileges, and runtime contexts shift.
The work gives a theoretical treatment of reliability via continual PAC-style learning: under explicit assumptions it derives an expected generalization bound, a retention bound tied to historical regression budgets, and a high-probability admission rule that requires fresh, independent validation data. It highlights organizational and safety consequences of persistent harness changes for knowledge sharing, scientific reproducibility, physical safety, and institutional accountability, and prescribes independent safeguards for learning, runtime privileges, and consequential actions. A comprehensive companion (code and supplements) expands on algorithms, evaluations, and broader impacts to guide practical, auditable deployment of evolving agent harnesses.
Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.
AI founders and executives in San Francisco discussed rapid revenue growth, data brokerage, and the evolving AI landscape, emphasizing the importance of data and model flexibility. They highlighted the potential of autonomous agents, the significance of data valuation, and ongoing reverse engineering efforts in the industry. (twitter.com)
Geoffrey Hinton warns that AI is creating alien-like beings and we have only a few years to adapt to this reality. He criticizes Donald Trump's understanding of superintelligence and emphasizes the need to prioritize human interests. (theguardian.com)
Anthropic suggests its AI may possess consciousness, raising ethical questions about AI feelings and suffering. The development has sparked debate among tech leaders about the moral implications and safety risks of advanced AI systems. (bbc.com)
Reinforcement learning in large language models can cause 'thinking leakage,' where models partially reactivate their original reasoning abilities rather than learning new behaviors. This leakage explains much of the observed performance improvements in models trained with RL in non-thinking modes. (arxiv.org)
Sergey Brin states that Google's AI models are entering a phase of recursive self-improvement, where AI helps develop better AI systems. This process could accelerate AI development significantly within the next year, according to industry experts. (businessinsider.com)
Linus Torvalds expressed a cautious stance on using AI for coding, emphasizing that understanding and effort are more important than the tools used. He believes people should do their best and understand their work regardless of the technology involved. (lore.kernel.org)
Today's best Hacker News stories, summarized and screenshotted, one email a day.