Tencent's CAFE: Self-Improving Search Agents Need Co-Evolving Feedback
Tencent-Hunyuan · hf · 2026-08-26
Tencent's Hunyuan team proposes CAFE, coupling a search agent and a critic via shared parameters to learn in-trajectory corrective feedback. The method reduces hallucinations and improves search performance across benchmarks.
More from coding & agent
- LangChain Founder: Deepagents Evolving into Multiplayer Harness — hwchase17 · 2026-08-26
- Set Max Runtime for Agents Using Pidot's Steer Function — viksit · 2026-08-26
- Ultimate cost-saving guide: Routing Claude Code via DeepSeek API — Unlikely_Bluejay5392 · 2026-08-26
- Practitioners Call for Open-Source Evals With Battle-Tested Infrastructure — samsja19 · 2026-08-26
- Developer automates house viewing schedule with a Tinder-style app — yacineMTB · 2026-08-26
- stealth-browser-mcp: undetectable browser sessions for MCP agents — tom_doerr · 2026-08-26