DFM · Discovery Foundation Models
AI has learned to answer questions. Next it learns to
We bring real scientific questions, data and experimental feedback into next-generation AI systems, on a path from assisting research to taking part in discovery.
A loop, not a line
Discovery does not run
from question to answer.
Latest workAll news →
- 19 Oct DiscoveryRSI, recursively learning to discover valuable hypotheses Details →
- 12 Oct RSI-Anything, world models for recursive self-improvement Details →
- 18 Sep JEPA-Anything, a cross-domain science world model Details →
- 17 Sep ScienceIDE, verifiable scientific environments Details →
- 16 Sep ScienceBuddy, the interactive research workspace Details →
- 15 Sep DFM technical report and research map Details →