Apr 3, 2026 · 29m · no-priors
AI for Atoms: How Periodic Labs is Revolutionizing Materials Engineering with Co-Founder Liam Fedus
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this episode of No Priors, Elad Gil interviews Periodic Labs co-founder Liam Fedus on transitioning from building landmark language models like ChatGPT to creating an AI foundation lab for physical matter. Fedus details how Periodic Labs integrates LLM orchestrators, symmetry-aware neural networks, and automated closed-loop experimentation to dramatically accelerate materials science and physical discovery.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. The hosts hold 32% of the talking time here. How this is scored →
The hosts as informed peer 5.7 Guest teaching 3.7 Guest disagreement 0.4 The hosts pushing back 0.4
speaking balance: gold is the hosts, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 23:42 Reframing recursive self-improvement timelines
Liam politely rejects Elad's timeline framing, pointing out that mastering self-improving code does not mean a model suddenly understands biology or scientific uncertainty.
Hardest push from the hosts ▶ 25:39 Challenging the need for advanced roboticsElad presses Liam on whether Periodic can realistically achieve escape velocity in closed-loop experimentation without breakthrough robotic systems like Physical Intelligence.
Biggest teaching moment ▶ 8:57 Literature noise and the necessity of closed-loop experimentsLiam explains why internet literature is insufficient for materials science, demonstrating that reported experimental values span orders of magnitude and require interactive physical loops.
The host holds their own ▶ 26:50 Elad detailing lab automation engineering at ColorElad demonstrates direct operational expertise by citing custom liquid handling adaptations, 3D printed vibration dampeners, and ML vision monitoring developed at Color.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | The hosts as informed peer | Guest teaching | Guest disagreement | The hosts pushing back | Why |
|---|---|---|---|---|---|---|
| Physics Roots and Why Physicists Excel in AI | 6 | 3 | 0 | 0 | Elad demonstrates industry awareness by listing prominent AI leaders with physics backgrounds like Dario Amodei and Adam Brown. Liam collaboratively explains the physics mindset and why post-Higgs physicists migrated to high-leverage AI research. | |
| Developing ChatGPT and Productizing GPT-4 at OpenAI | 4 | 3 | 0 | 0 | Liam shares the insider history of productizing GPT-4 into ChatGPT under John Schulman's direction. Elad asks about the pivot from language models to physical atoms, which Liam explains as the necessary next frontier for real scientific acceleration. | |
| Solving the Physical Data Gap with Closed-Loop Experiments | 5 | 6 | 1 | 1 | Elad inquires how Periodic overcomes the physical data bottleneck compared to internet-scale LLM pre-training. Liam educates on why scraped literature data is flawed due to values spanning orders of magnitude, highlighting the necessity of closed-loop experimental grounding. | |
| Generalization Across Quantum Domains vs AlphaFold Paradigm | 7 | 4 | 0 | 1 | Elad leverages his biology background in X-ray crystallography and NMR to question domain generalization versus AlphaFold's structural dataset. Liam clarifies how quantum mechanical representations generalize across chemistry but decouple at higher abstraction layers like fluid dynamics. | |
| Commercialization: Software Intelligence Layer vs Discovery Model | 6 | 3 | 0 | 0 | Elad explores business models, contrasting broad language interfaces with vertical discovery and biotech royalty structures. Liam explains that Periodic operates primarily as a software intelligence layer and control plane rather than a pure discovery play. | |
| The Diamond Age and the Ten-Year Vision for Physical Matter | 6 | 2 | 0 | 0 | Elad references Neal Stephenson's science fiction novel The Diamond Age to frame the long-term vision of matter generation. Liam outlines his 10-year vision for accelerating atomic synthesis to match digital development speeds. | |
| Multidisciplinary Collaboration and Bringing Scaling Laws to Science | 5 | 3 | 0 | 0 | Liam details how bringing scaling laws and industrial automation to multidisciplinary teams transforms physical research. Elad notes the stark economic discrepancy between academic postdocs and machine learning engineers. | |
| Spiky Intelligence, Domain Gaps, and Recursive Self-Improvement | 5 | 6 | 2 | 1 | Liam reframes Elad's question about generalized self-improvement timelines by rejecting the concept of intelligence as a single scalar. He highlights how verifiable closed loops in software engineering do not trivially translate to decision-making under uncertainty in physical sciences. | |
| The Role of Robotics in Lab Automation and Closed Loops | 7 | 3 | 1 | 1 | Elad brings up deep technical experience from his company Color, explaining custom liquid handling robotics, vibration reduction, and ML monitoring. Liam agrees that general robotics will accelerate setup times while clarifying that Periodic currently succeeds with standard automation. |
Statements from this episode (17)
Fedus: Post-Higgs experimental bottlenecks drove physicists to transition into AI
“After the discovery of the Higgs I think a lot of high energy physicists were sort of looking for what's next. Ultimately it becomes bottlenecked on the new apparatus for, you know, pushing the next energy frontier. And I think a lot of physicists were looking…”
Fedus: John Schulman steered OpenAI toward a general chatbot over narrow tools
“And we're all spitballing ideas, like writing bot coding bot. You know, very natural at the time. Some of our least interesting ideas were a meeting bot, so it would just sit in a Google Meet, take notes, and then send out, like, to-dos after. But John Schulma…”
Fedus: AI must connect to the physical world to accelerate science
“The opinion that I and others held as periodic was, you're not going to see the same kind of acceleration in science and technology unless you start connecting these things to the physical world.”
Fedus: Reasoning and coding agents connect software AI to physical domains
“And I think those were foundational technologies necessary To then connect these systems to the physical world. Like it was just not impossible, not possible with like the AI technology of.”
Periodic Labs spends zero effort building or fine-tuning coding models
“Periodic spends zero effort on improving coding models. We're, you know, incredibly impressed by codex, cloud code. And so that's been a huge accelerator for the company. But focus our machine learning efforts where You know, the existing frontiers is not suff…”
Fedus: Training materials AI on academic literature fails to find ground truth
“One of the engineers on our team was looking at a reported material property and It was just sort of extracted values from literature, and it was really interesting to see the reported value spanned many orders of magnitude. And so you train an ML system on th…”
Fedus: Quantum AI models do not generalize across abstraction levels to fluids
“But like if you produce a system that has modeled quantum mechanical objects really accurately, it's not really helping much on like, you know, fluid dynamics or, you know, like another kind of like level of abstraction.”
Periodic Labs uses LLMs to orchestrate symmetry-aware atomic neural networks
“We think about them almost as like an orchestration layer. So that's sort of a co-pilot assistant, but also like a system that can direct experiments. And it's almost, it's orchestrating other specialized models as well. So we do construct neural nets that are…”
Fedus: Periodic Labs launches as a software intelligence layer, not discovery model
“We're thinking about us ourselves as an intelligence layer for these companies. So you can think about system of record control plane for different Experiments and getting to solutions. But like you're saying, there is a very interesting aspect of some breakth…”
Fedus: AI Matter Synthesis Will Profoundly Impact Semiconductors, Aerospace, and Energy
“As you're pointing out, you're going from systems that aren't just writing essays, not just writing software, but to literally generating matter. And I think it has pretty profound implications to semiconductors, airspace, energy.”
Fedus: Physical sciences and engineering will follow machine learning scaling laws
“And I think the physical sciences, physical engineering, Will have a very similar property where we establish these scaling properties and Bring that mindset.”
Fedus: Compute costs outweigh physical infrastructure in materials AI labs
“What's interesting is just the compute costs relative to physical infrastructure is actually surprising where, you know, so much money is spent on the compute that the physical infrastructure sometimes is actually lower, but, you know, has very large lead time…”
Fedus: Intelligence is not scalar; AI shows extreme, non-intuitive spikiness
“I think one fallacy is thinking about intelligence as a scalar. We've consistently seen these systems have a very odd spikiness, and it's actually possible to architect a system that is world-class on some math domain, but then you could do some perturbations …”
Fedus: Self-improving software AI will not automatically generalize to domains like biology
“So rolling forward that software engineering self-improvement, I think you're going to have a system that can write complete repositories, identify bugs, refactor code, But it doesn't suddenly understand biology. Right? It's just like there's a domain gap ther…”
Fedus: Recursive self-improvement is happening in software and entering AI research
“So, In that domain, I think it's happening now-ish. And I think we'll see the same thing too for AI research. That's a slower outer loop because now the experiment isn't just checking some unit tests passing, but it's checking what was the scaling property? Di…”