Unverified50% confidenceFactExact time
ROMA delegates the three decisions of 'when to explore, what to explore, and when to stop' entirely to language model reasoning, rather than hard-coding them as fixed rules.
1
Sources
50%
Confidence
Long-term
Relevance
10/10/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedAt inference time, the model uses a dynamic threshold of mean plus standard deviation to filter sentences, with a top-3 fallback mechanism to ensure at least three sentences are selected.63% similarUnverifiedSTRIPS定义了用「前置条件—动作—后置条件」三元组描述规划问题的语言62% similarUnverified该案例对能精确匹配的字段使用规则判断，对需要语义理解的内容交由模型处理，形成规则+语义的混合判断模式62% similarUnverifiedCodex的计划模式机制理论基础来自大语言模型领域的思维链推理（Chain-of-Thought Reasoning）62% similarUnverifiedFor knowledge points with multiple examples (such as three ways to get structured output from a model), it is sufficient to thoroughly implement only the most important one and understand the principles of the rest.61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/994284API
curl https://kongchang.com/api/v1/knowledge/claims/994284MCP
get_claim(id=994284)