Unverified50% confidenceFactExact time
After the agent completed the map, the authors further improved the policy through an AlphaZero-style expert loop: combining search with distillation, distilling the optimal actions obtained from search into the policy network.
1
Sources
50%
Confidence
Long-term
Relevance
10/24/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe intelligence of the search experience—understanding intent, multi-step retrieval, result synthesis—mainly happens at the model and orchestration layers, not at the protocol layer.68% similarUnverifiedValerant adopts an exploration-driven action selection strategy, actively choosing actions that expand unknown regions and complete the map.68% similarUnverifiedCompeting products such as Anthropic's Claude and Google's Gemini are continuously catching up to OpenAI in tool-calling and agent capabilities.66% similarUnverifiedThe research agent in Anthropic's official demo repository is coordinated by a lead agent, which splits subtopics and dispatches multiple researcher sub-agents in parallel to search the web66% similarUnverifiedThe project represents a new paradigm of Expert Behavior Mining—automatically extracting patterns and principles from traces of expert historical behavior.65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1094080API
curl https://kongchang.com/api/v1/knowledge/claims/1094080MCP
get_claim(id=1094080)