What we are building, what the experiments show, and the questions that come next.
Shared evidence, independent role and coordination edits, and a new memory-policy editor. Updated results with GPT-6-Astra and GLM-5.3.
Read the story & watch the film →Can an LLM program a genomic model without gradient descent? Direct weight edits, working circuits, and lessons from 10 GUE tasks.
Read the story & watch the video →How black-box rewards steer discrete diffusion toward useful DNA and protein sequences, without retraining. Includes our 68-second research film.
Read the story & watch the film →
