Unverified50% confidenceFactExact time
The seven LLMs had a low success rate on the task of discovering the underlying expression, and their intrinsic inquiry ability is quite limited.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/16/2026
First Seen
Valid until: 1/14/2027
Sources
Related Entities
Related Claims
UnverifiedA lack of information-gain awareness leads to low sample efficiency in LLMs; with the same number of interactions, human inquirers can pin down the underlying rules faster.72% similarUnverifiedSparse rewards only provide signals when a task succeeds or fails, which leads to extremely low exploration efficiency in long-horizon, multi-stage tasks.72% similarUnverifiedThe study observed an inherent difficulty gap across different tasks, with LLMs performing notably weaker on certain types of reasoning tasks.71% similarUnverifiedIn this task, scaling the model from 7B/8B to 14B yielded only marginal performance gains.71% similarVerifiedThe limitation of specialized small models is that their generalization ability rapidly degrades when task boundaries are ambiguous, cross-domain reasoning is required, or unseen edge cases need to be handled.70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1047436API
curl https://kongchang.com/api/v1/knowledge/claims/1047436MCP
get_claim(id=1047436)