This is really cool.

PhAI Labs just released DFM (Discovery Foundation Models), a research paradigm for AI systems meant to participate in scientific discovery.

Most AI-for-science systems start after the hard choices have already been made.

The question is defined. The task boundary exists. The model is asked to reason, search, or optimize inside it.

But DFM (Discovery Foundation Models) targets that layer.

The loop begins with a partially understood world, moves through a valuable unknown and a researchable question, then into explanation, intervention, and externally verified knowledge.

PhAI Labs is opening a Scientist Collaboration Program so researchers can bring real problems

So scientists bring the question, data, experimental conditions, and expert judgment. The AI side brings models, tools, analysis, and a way to keep revising the investigation as evidence arrives.