Unverified50% confidenceFactExact time
The paper's failure analysis annotated 644 failed attempts, of which 93.3% were faithful implementations of modifications that brought no improvement, and only 6.7% were implementation or execution problems.
1
Sources
50%
Confidence
Long-term
Relevance
10/20/2026
First Seen
Sources
SwarmRSI: How a Swarm of Research Agents Achieves Recursive Self-Improvement
bilibili熊二等兵10/18/2026
Related Entities
Related Claims
UnverifiedReplay-based validation can detect 76.2% of unsuccessful improvements, whereas build-check-only approaches fail to detect any of these behavioral failures.80% similarUnverifiedAmong the 35 genuine failure cases, only 3 did not involve any assertions, indicating that the vast majority of failures stem from accepting biased statements rather than from missing information.74% similarUnverified一个 90% 成功、10% 出错的工具调用系统几乎无法投入实际使用，因为单次错误会污染后续上下文造成级联失败72% similarUnverifiedIn tasks with propositional constraints, the proportion of models integrating candidates already verified as invalid is as high as 93%–100%.72% similarUnverifiedIn a test of 119 pairs of "near-identical questions", at the 0.81 threshold 62% of modified questions incorrectly hit; even raised to 0.90, 25% still hit; the hit rate for genuine synonymous rewrites dropped from 39% to 4%.72% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1073056API
curl https://kongchang.com/api/v1/knowledge/claims/1073056MCP
get_claim(id=1073056)