Meta's MIMESIS: A 9B User Simulator That Beats Claude Opus 5 on Behavioral Fidelity
dair_ai · x · 2026-10-09
Meta Superintelligence Labs published a paper on user simulators for agent training, identifying a common flaw in agent RL setups.
- Assistant LLMs usually play the user, making simulated users too cooperative and explicit; a fixed GPT-5.5 agent finds tau-bench tasks easier with these fake users than with real people
- The team trained MIMESIS, a 9B user simulator, on human conversations and 13 real-world user behavior patterns — beating Claude Opus 5 on behavioral fidelity by 13.4 points
- Agents trained against MIMESIS outperform agents trained against GPT-5.5 across all nine unseen user simulators
- A coaching step that converts the simulator's private reasoning into feedback adds further gains
Related event: Meta open-sources MIMESIS user simulator to boost agent training(2 posts)→
More from coding & agent
- DuckDB v2.0 CLI agent mode cuts agent-read tokens by 59% on TPC-H benchmarks — josh_wills · 2026-10-10
- Claude Code agents beat torch.compile on CUDA kernels: 2.5x fused GELU, 1.57x matmul via fp32-splitting trick — lmoroney · 2026-10-10
- Roadie: open-source Go USB KVM gives AI agents hands via HTTP — hugs · 2026-10-10
- VirusTotal: fastest-growing AI agent ecosystem OpenClaw becomes a malware delivery channel — Bedrovelsen · 2026-10-10
- Nuwa.skill hits 33.9k GitHub stars distilling anyone's thinking into agent skills — AlchainHust · 2026-10-10
- Microsoft launches Database Hub in Fabric: one place to detect, investigate and automate database ops — adnan_hashmi · 2026-10-10