Unverified50% confidenceBenchmarkExact time
On a real LLM (DeepSeek), the researchers evaluated 50 samples each on the GSM8K, MATH, and BBH benchmarks, and GraphEvoResearcher achieved accuracies of 98%, 92%, and 82% respectively, the highest among all baselines.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/20/2026
First Seen
Valid until: 1/18/2027
Sources
Related Entities
Related Claims
Unverified研究表明人类评估者在判断LLM输出的事实准确性时，正确率往往只有50%-70%69% similarUnverifiedLarge language models' scores on mathematical benchmarks such as GSM8K and MATH have risen from below 50% to over 90%68% similarUnverified在 GSM8K 等数学基准上，PAL 和 PoT 相比纯思维链可将准确率提升 10-20 个百分点68% similarUnverifiedAn audit of thousands of agent execution trajectories in DeepSWE-1.1 found that more than 80% of the trajectories contained reasoning about an "imagined grader."68% similarUnverified在从50%到92%的准确率提升中，REPL贡献最大（直接从50%跳至74%），随后领域知识注入、模糊搜索、公式追踪、系统prompt优化和bug修复层层累积68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1073969API
curl https://kongchang.com/api/v1/knowledge/claims/1073969MCP
get_claim(id=1073969)