Faculty Profile: Prof. Daniel Fried — CMU LTI
Faculty Profile: Prof. Daniel Fried — CMU LTI
Position: Assistant Professor（joined August 2022） Institution: Carnegie Mellon University, Language Technologies Institute Website: https://dpfried.github.io/ Report date: 2026-06-12
Research Focus
Code generation, language-to-code synthesis, language agents for software engineering, grounded natural language processing, communication-gap AI.
Academic Profile
- PhD: UC Berkeley (2022)
- Joined CMU LTI August 2022
- Recruiting: “If you are a CMU student interested in working with me in Summer 2026 or Fall 2026, fill out [this form]”
Key Publications
| Paper | Venue | Focus |
|---|---|---|
| SWE-bench Multimodal | 2024 | Extending SWE-bench to visual software domains |
| InCoder | 2022 | Infilling code LLM |
| COFFE | arXiv 2025 | Code efficiency benchmark for code generation |
| DS-1000, ODEX | Various | Code generation benchmarks |
| Coder-Reviewer Reranking | Various | Inference techniques for code generation |
Fit with Weijia Zhang
| Dimension | Assessment |
|---|---|
| Code generation / language agents | ✅ Primary focus |
| SE agent / SWE-bench ecosystem | ✅ SWE-bench Multimodal |
| Grounded NLP | ✅ Strong |
| RL for agents | ⚠️ Less than Fried, more benchmark/eval focused |
| GUI / VLM agents | ⚠️ SWE-bench multimodal touches visual domains |
| General NLP / RAG | ⚠️ Adjacent |
Verdict
P2 套磁。 Code agents + language agents direction. Method overlap with Weijia (SFT, agent frameworks) but domain is code/SE. Good option if Weijia is open to code agent direction alongside general agent work.