Survey frames risks of recursive self-improvement in large-model systems
Original title: Trustworthy Recursive Self-Improvement in the Era of Large Models: Foundation, Risks, Evaluation and Open Challenges
The authors define RSI and organize trustworthy improvement around validity, constraints, security, and oversight.
A survey defines recursive self-improvement as persistent changes that can also revise the mechanisms governing future updates. It argues self-modification does not ensure genuine gains, preserved constraints, or lasting verification. Risks include flawed evaluations, capability loss, weaker compliance, persistence of compromised artifacts, and reduced human oversight. The paper reviews safeguards and evaluation evidence and lists requirements for independent, longitudinal assessment.
Unconfirmed: The paper is a survey and does not report new experimental measurements.
- Four themes cover improvement, alignment, security and privacy, and controllability.
- The analysis spans diagnosis, proposal generation, evaluation, and update implementation.
- The survey separates direct RSI findings from evidence in related systems.
- It calls for independent and longitudinal evaluation of successor systems.
From the original article
1 min read at arxiv.orgAI systems built around large models increasingly use experience and feedback to make persistent changes to models, prompts, memories, tools, and workflows. Recursive self-improvement (RSI) extends this process to inductive biases and to the mechanisms that govern future updates, which can themselves be revised and reused across improvement cycles.
By Jiayue Pu · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.