In 12 weeks, we built a research facility that is run entirely by AI.
Summary
C5R CORP built an AI-run research facility in 12 weeks that uses GPT-6Astra to design, execute, and observe experiments across biology, chemistry, and materials science.
Hardware integration enables AI control.
The facility uses an ontology of actions and an inventory system to manage 40+ scientific instruments through reverse-engineered firmware and custom adapters.
integrated 40+ scientific instruments – by reverse engineering drivers
Models design experiments as code.
Given a goal, models explore inventory and read specifications to generate code that converts into equipment control and instructions for human personnel.
models can explore inventory and read specs to design experiments as code
Physicality limits current model performance.
While models possess theoretical knowledge, they fail at physical laboratory tasks like accounting for evaporating solvents or avoiding DNA contamination.
They pipette frozen samples, contaminate DNA, fail to account for evaporating solvents
Science lacks instant verification loops.
Unlike software or mathematics, science is an intrinsically physical endeavor that lacks the immediate feedback cycles available to digital models.
Models doing Software and Mathematics have had the benefit of instant verification loops
SciUniverse benchmarks real-world research.
The SciUniverse benchmark evaluates the entire research process, including protocol debugging, instrument operation, and long-horizon tasks.
spans the research process including protocol debugging and bring-up
Read the original ↗Save to your Sponge
Summarized with Sponge. Quotes are the author’s own words.