Episode summary
Building an AI Physicist: ChatGPT Co-Creator’s Next Venture
The a16z Show
· · Hosts: Anjney Midha · Guests: Liam Fedus, Ekin Dogus Cubuk
Summary
Periodic Labs is trying to build what the guests call an “AI physicist”: a system that can use LLMs, simulations, and real experiments together to accelerate discovery in physics, chemistry, and materials science. Their core argument is that language-model scaling alone is not enough for hard science, because the relevant reward signal is not internet text or human preference, but whether an experiment in the real world works. They want to turn nature itself into the feedback loop.
The guests explain that this starts with domains like solid-state chemistry, superconductivity, and magnetism, where outcomes are measurable and the lab can generate high-throughput, high-quality data. They say the first big benchmark is whether Periodic can push beyond today’s roughly 135 Kelvin ambient-pressure superconductors, and eventually toward a 200 Kelvin superconductor. They also argue that the company’s models will need mid-training on experimental and simulation data, plus RL loops that reward scientific reasoning, not just text prediction.
They frame the startup as both a research program and a commercial platform for advanced industries like semiconductors, defense, space, and manufacturing. The model should become an intelligence layer for teams with large R&D budgets, helping them automate simulations, preserve expertise, and shorten iteration cycles.
Key takeaways
- The guests say Periodic Labs is building an AI system that learns physics and chemistry from experiments, not just text, because real-world verification is the reward signal.
- They view superconductivity as the best first target, with today’s ambient-pressure benchmark around 135 Kelvin and a longer-term aspiration of 200 Kelvin.
- The company plans to use mid-training on experimental and simulation data to make models more useful for scientific reasoning and materials discovery.
- They think the first commercial path is helping advanced industries like semiconductors, defense, space, and manufacturing accelerate R&D and simulations.
- They argue that scaling laws still matter, but the slope on out-of-domain scientific tasks is often too small without domain-specific data and feedback loops.
Notable quotes
“ultimately sciences driven against experiment in the real world. And so that's what we're doing with periodic labs”
“Nature like is our RL environment in in our setting”
“we really want to see technology out in the world and we felt like this was just the right place to begin”
“What is the highest temperature superconductor we synthesized? Uh today the best number for ambient pressure is 135 Kelvin or so”
How the episode opens
Ultimately sciences driven against experiment in the real world. And so that's what we're doing with periodic labs. We're taking these precursor technologies and we're saying okay if you care about advancing science we need to have experiment in the loop.
The applications of building an AI physicist for lack of a better word that can design the real world are so broad. You can apply them to advanced manufacturing. You can apply them to material science to chemistry any process where there's R&D with the physical world required.
It seems like we'll benefit from breakthroughs that periodic is working on. For example, if we could find a 200 Kelvin superconductor even before we make any product with it. To be able to see such quantum effects on at such high temperatures, I think would be such an update to people's view of how they see the universe.
So Liam, you were the co-creator of Chad Gupt. You were running some of the physics teams at Deep Mind. Let's talk about how you guys met and what was the moment where you realized that you guys had to leave both of those labs to start periodic.
I believe we met eight years ago at Google brain flipping over a large tire. Y at the Google they got to give us more on that story. Um, so Google Rails, uh, was one of the gyms at Google at the Google facilities and I I think that's where Doge and I met and it was just this massive tire that like a single person basically can't flip single like by themselves.
And so Doge was trying to flip it and he like pulled me over. He's like, I think the two of us could do it. So, and why were you trying to flip this tire?
From the episode’s captions: roughly the first two minutes, lightly cleaned. Keep watching on YouTube →
Topics
- llms in physics and chemistry
- scaling laws and model verification
- automating discovery in hard sciences
- superconductivity
- commercializing ai research
Companies and tickers mentioned
- GOOGL
- GOOG
- a16z
- OpenAI
- Google DeepMind
- Periodic Labs
- Periodic
Watch the episode
Watch “Building an AI Physicist: ChatGPT Co-Creator’s Next Venture” on YouTube
Summary, takeaways and quote selection by Octagon. Quotes are short excerpts; the full conversation is in the episode.
More from The a16z Show
-
How AI Could Bring Halo and Other Classic Games Back to Life -
Mike Volpi on Why AI Infrastructure Will Outlast the Hype Cycle -
Why HubSpot’s Co-Founder Is Betting Against the SaaSpocalypse -
Inside Salesforce’s Acquisition Strategy for the AI Era -
How AI Is Turning Figma’s Finance Team Into Builders -
The New AI Agents That Shop, Plan, and Gossip
The a16z Show, this episode and its artwork belong to Andreessen Horowitz. Octagon is not affiliated with the show; this page is an independent summary.
