The authors trained MIMESIS using human conversations and human-simulation tasks, followed by supervised fine-tuning on ThoughtTrace.
Realistic simulated users improve training for interactive language agents
MIMESIS models varied human behavior and provides private reasoning signals that help agents generalize across simulated users.
Academic
Hoang Phan · Dat Huynh · Andrey Zhmoginov · Qi Zeng · Wancen Mu · Yue Cao · +4 more
Meta Superintelligence Labs · New York University · University of Wisconsin - Madison
Research Digest··3 min read
Phan et al.
Why this paper
From Meta Superintelligence Labs and 2 others
In one line
MIMESIS learns realistic user behaviors from human conversations and reasoning traces, enabling agents to train more effectively and generalize across unseen users.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ✓Compute or model size stated (params 9B)
- ·No stated limitations found
- ✓Reports numbers on named benchmarks (3 benchmarks)
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§