Unverified50% confidenceBenchmarkExact time
In cross-task transfer experiments, the research states inherited from three source tasks raised the average base value of 12 held-out NatureBench tasks from 12.01 under cold start to 21.89, and the proportion matching SOTA jumped from 41.7% to 86.1%.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/20/2026
First Seen
Valid until: 1/18/2027
Sources
SwarmRSI: How a Swarm of Research Agents Achieves Recursive Self-Improvement
bilibili熊二等兵10/18/2026
Related Entities
Related Claims
UnverifiedIn the ScopeBench evaluation, the raw capability of the 8 models ranged from 12.2% to 81.1%, and the scope compliance capability ranged from 34.4% to 86.7%.63% similarUnverifiedBecause the experiment used temperature=0 and a fixed seed in a single deterministic run, any field flip moves the score by about 2.8 percentage points out of 12 tasks × 3 fields = 36 units, so the gap between 91.7% and 83.3% cannot be directly generalized into a statistically meaningful general advantage.61% similarUnverifiedAmong the 12 model × task combinations, only gpt-5.4-mini showed a statistically significant accuracy improvement on the logical reasoning task: jumping from 15% to 97.5%, with p=0.00056 after Holm correction.60% similarUnverifiedAmong the 90 tasks in NatureBench, 77% of the final best programs came from at least one recombination, crossover or feedback-triggered update.60% similarUnverifiedIn a natural experiment, 7 of 11 skill assessment scores changed between two runs, with a maximum fluctuation of 1 point.60% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1073054API
curl https://kongchang.com/api/v1/knowledge/claims/1073054MCP
get_claim(id=1073054)