Unverified50% confidenceFactExact time
In the test-time planning setting, the agent uses the world model to search online for action sequences, and its generalization ability is usually better than that of direct policy learning.
1
Sources
50%
Confidence
Long-term
Relevance
11/2/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedTest-time search refers to the AI running a search process in real time during the inference phase of actual gameplay (such as Monte Carlo Tree Search) to find better moves, which in hidden-information scenarios requires simultaneously traversing multiple possible opponent state combinations75% similarUnverifiedResearch shows that effective search strategies can outperform raw model scale.74% similarUnverifiedGenerative retrieval allows models to directly generate document identifiers or answers, while agentic retrieval endows systems with the ability to autonomously plan, conduct multi-round retrieval, and evaluate result quality.72% similarUnverifiedWhen dealing with highly time-sensitive information, models with internet search or reasoning capabilities should be prioritized.72% similarUnverifiedCommon hyperparameter tuning strategies include grid search, random search, and Bayesian optimization; the latter models historical evaluation results to guide the next sampling and typically achieves better results with fewer trials.72% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1126107API
curl https://kongchang.com/api/v1/knowledge/claims/1126107MCP
get_claim(id=1126107)