Faculty Profile: Prof. Maarten Sap — CMU LTI
Faculty Profile: Prof. Maarten Sap — CMU LTI
Position: Assistant Professor Institution: Carnegie Mellon University, Language Technologies Institute Website: https://maartensap.com/ Report date: 2026-06-12
Research Focus
Social NLP, LLM social intelligence, agent social evaluation (SOTOPIA), AI truthfulness/deception, societal implications of AI, bias and safety in LLMs.
Academic Profile
- PhD: University of Washington (2021)
- 2025 Packard Fellow
- 2025 Okawa Research Grant Award
- Amazon Research Award (Generative AI)
- NSF CAREER (implied by Packard fellowship eligibility timeline)
Key Publications
| Paper | Venue | Focus |
|---|---|---|
| SOTOPIA | ICLR’24 | Interactive evaluation framework for social intelligence in language agents |
| SOTOPIA-π | ACL’24 | Interactive learning of socially intelligent language agents |
| SOTOPIA-S⁴ | arXiv 2025 | Flexible, customizable large-scale social simulation |
| AI-LieDar | NAACL’25 | Trade-off between utility and truthfulness in LLM agents |
Fit with Weijia Zhang
| Dimension | Assessment |
|---|---|
| Agent evaluation frameworks | ✅ SOTOPIA is a major contribution to agent eval |
| Social AI / multi-agent interaction | ✅ Core |
| LLM truthfulness / alignment | ✅ Strong |
| General NLP agents | ✅ Adjacent |
| RL post-training | ❌ Not his primary focus |
| GUI / VLM agents | ❌ Text-based social simulation |
Verdict
P2 套磁。 Relevant if Weijia’s work touches agent evaluation, multi-agent social interaction, or alignment. SOTOPIA framework is directly usable for evaluating agents in interactive settings. Less relevant for RL training / GUI agent direction.