Unverified50% confidenceFactExact time
FOCUS conducts experiments on three types of benchmarks — legal reasoning, deep search, and code localization — comparing against GRPO and VinePPO, and outperforms both.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/27/2026
First Seen
Valid until: 1/25/2027
Sources
Related Entities
Related Claims
UnverifiedThe experiment compared four strategies—direct prompting, supervised fine-tuning, chain-of-thought reasoning, and rule generation under GRPO training—and GRPO combined with rule generation achieved the best performance on all four ToM benchmarks.69% similarUnverifiedThe BRIGHT benchmark specifically evaluates how retrieval systems perform in scenarios requiring multi-step reasoning, with typical tasks including math problem retrieval, code snippet matching and scientific reasoning document localization.69% similarUnverifiedThe core idea of the retrieve-and-verify framework derives from retrieval-augmented generation (RAG), first retrieving relevant segments and then performing verificatory reasoning.67% similarUnverified现有基准GAIA、BrowseComp只提供明确查询，仅考验多步推理能力，未考虑歧义问题67% similarUnverifiedThe evaluation adopts progressively stricter nested matching criteria, encompassing three dimensions: concept hit, evidence snippet localization, and correctness judgment.66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1102744API
curl https://kongchang.com/api/v1/knowledge/claims/1102744MCP
get_claim(id=1102744)