Unverified50% confidenceFactExact time
CoReMem's reward design prioritizes problem-solving rate while penalizing the number of steps in successful runs.
1
Sources
50%
Confidence
Long-term
Relevance
11/13/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedMeta-reward here does not refer to the correctness score of the task itself but to an evaluation signal of "reasoning process quality," used to decide the next meta-action such as revise, backtrack, or stop.67% similarUnverifiedProcess supervision refers to giving feedback signals on intermediate steps of reasoning or interaction rather than judging only the final outcome; its advantage is providing denser learning signals and alleviating the long-horizon credit assignment problem, while its drawbacks are high annotation cost, potentially inaccurate process rewards, and the model's tendency to overfit to seemingly correct intermediate steps.67% similarUnverifiedThe method depends on the quality of question generation, and its generalization ability across different tasks still needs further verification.65% similarUnverifiedHumans solving problems usually proceed in two steps: first generate several possible solutions (generation), then evaluate which is most suitable (selection); LLMs are good at generation, but the selection step must be done by humans.65% similarUnverifiedFeasible engineering practices include setting explicit success/failure criteria for the task, estimating the marginal benefit of each retry, setting a total action budget, and recording failure reasons for review.64% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1166040API
curl https://kongchang.com/api/v1/knowledge/claims/1166040MCP
get_claim(id=1166040)