Carl Edwards

AI Scientist @Genentech. CS PhD @UIUC. Scientific NLP, AI4Science, biology FMs, drug discovery.

[Carl Edwards @CarlEdwards_NLP]

Check it out here:

🌐 Website: https://t.co/rFvmtPF8kw
🤗 Dataset: https://t.co/nba7KGJFx5
💻 Code: https://t.co/nAjVnPJMmT
📕 Paper: https://t.co/g2pMGYJ0aF

[Carl Edwards @CarlEdwards_NLP]

Excited to announce that AssayBench has been accepted at the NeurIPS E&D track 🎉🎉!!
See below ↓ a video explaining how it works!

[Carl Edwards @CarlEdwards_NLP]

Just realized I put the wrong link for trying the model 🫠.

https://t.co/OCjqvHfXY2 is the right one.

[Carl Edwards @CarlEdwards_NLP]

Autonomous science is getting closer: systems that run experiments, learn from results, and choose what to test next.
But can they learn how to experiment from existing experiments? In AssayLoop, we train across CRISPR screens to do that.

Try it: https://t.co/bx6YnBY2Cg 🧵

[Carl Edwards @CarlEdwards_NLP]

A slightly philosophical point: because experiments are specified in natural language, in the limit a sufficiently capable model would have knowledge that goes beyond what is currently known by human scientists!

[Carl Edwards @CarlEdwards_NLP]

📃 Paper: https://t.co/iw7S53NV1V
🌐 Website: https://t.co/lx9G6bsXFB
💻 Code: https://t.co/xTB2T7YoU9
🤗 Data/models: https://t.co/neaDugWZzv

[Carl Edwards @CarlEdwards_NLP]

Organized and sponsored by @Genentech. Submission deadline: July 22, 2026.

Get started → https://t.co/9gn2luoLnY

Three Kaggle leaderboards:
A: https://t.co/HyM1BSWybg
B: https://t.co/x0DBaLefrs
C: https://t.co/aHpKdFmJmQ

[Carl Edwards @CarlEdwards_NLP]

Can LLMs and agents do real biology, or just talk about it?

We're launching the BioReasoning Challenge 2026 at MLGenX @ #ICLR2026: an open competition asking whether language models can predict the outcome of a real perturbation experiment.

https://t.co/0W7p6RY1Tg

[Carl Edwards @CarlEdwards_NLP]

Three tracks, three research questions:

A. Prompt-only: find the best prompt for a fixed GPT-OSS-120B model
B. Agentic: multi-agent + tool calls, same fixed LLM
C. Fine-tuning: any open model <10B params (SFT, LoRA, RL, ...)

Submit to one or all.

[Qingyun Wang @eagle_hz]

🚀Join us tomorrow in Room 122A for our 2nd AAAI @RealAAAI AI4Research workshop🧪! We have an exciting lineup of presentations from our amazing invited speakers—don’t miss out! #AI4Research #AAAI2025 https://t.co/462jpnN9iK