01 The AI scientist
Intelligence, held to the test of nature.
Rafati builds an AI scientist together with the autonomous laboratories where it works. Software forms a hypothesis, runs the real experiment, and learns from what the physical world returns. Knowledge is not something you read. It is something you make.
02 The thesis
New knowledge is made only when an idea meets the resistance of the real world.
The last decade of scientific AI leaned almost entirely on models trained on the text of the internet, a finite resource that the leading systems have now largely exhausted. The discoveries ahead will not be found on a page. They have to be produced, one experiment at a time, and read back from nature itself.
A single experiment can return gigabytes of signal that exists in no paper, dataset, or archive. Our models read all of it.
Automated facilities run without a queue or idle bench time, compounding the dataset day after day.
Results thrown away. Every outcome is kept, including the negative results the literature almost never publishes.
03 The premise
Intelligence alone does not move science forward.
A model can reason beautifully and still discover nothing. Reasoning rearranges what is already known. Discovery requires contact with a world that can surprise you, and that is precisely what a language model, on its own, has never had.
So we do not stop at building smarter models. We build the places where those models can act: on real instruments, real materials, and real results. The model proposes, the laboratory performs, and reality decides. That verdict, kept and studied in full, becomes the teacher.
04 What we are building
Three pieces, engineered to work as one.
The model, the laboratories, and the loop that joins them. None of the three is useful alone. Together they turn a question into an answer, and an answer into the next, sharper question.
The model
Reasons and proposes
It turns a goal into a testable prediction, designs the experiment that will settle it, and updates its beliefs the moment results arrive.
Read the approach
02The laboratories
Act on the world
Robotic handling and precision sensing carry out experiments on real materials, continuously, and measure them down to the gigabyte.
Tour the labs
03The loop
Joins the two
Hypothesis, experiment, measurement, revision, closed and repeated with physical reality as the only judge that counts.
See the loop
05 The discovery loop
How discovery happens here.
One full turn of the engine, in three movements. The model designs, the lab runs, and reality answers. Then the answer becomes the next question, faster than the turn before.
Hypothesize
Experiment
Learn
07 Work with us
Let's discover something together.
If you are a scientist, an engineer, an industry partner, or a builder who wants ideas to meet reality, we would like to hear from you.