scripod.com

Building an AI Physicist: ChatGPT Co-Creator’s Next Venture

The a16z Show

2025/09/30
The a16z Show

The a16z Show

2025/09/30

Shownote

Scaling laws took us from GPT-1 to GPT-5 Pro. But in order to crack physics, we’ll need a different approach.  In this episode, a16z General Partner Anjney Midha talks to Liam Fedus, former VP of post-training research and co-creator of ChatGPT at OpenAI,...

Highlights

The pursuit of scientific breakthroughs has long been constrained by the pace of experimentation and theoretical exploration. Now, a new wave of AI-driven research is emerging—one that doesn't just analyze data but actively participates in discovery. In this conversation, two pioneers from leading AI labs unveil their vision for a system that learns not from text alone, but from the physical world itself.
06:21
The next frontier in AI is physics, not just logic or math.
09:27
Teaching LLMs scientific inquiry through real quantum-level experiments is essential to reduce epistemic uncertainty.
17:59
A model trained on coding data won't cure cancer without relevant scientific data
24:19
High-temperature superconductivity unites physicists and offers profound technical and philosophical value.
33:21
A great researcher at Periodic Labs is highly mission-driven, focused on accelerating science.
42:14
Models are currently poor at scientific analysis but can improve through mid-training and data integration.
48:36
Periodic Labs is launching a grant program to support academic research in LLMs and materials discovery.

Chapters

What happens when AI learns from real experiments, not just text?
00:00
Can AI truly understand quantum mechanics through physical interaction?
09:27
Why general-purpose models fall short in scientific discovery
17:59
The moonshot goal: cracking high-temperature superconductivity
24:19
Bridging the gap between AI researchers and physical scientists
30:40
How to teach AI new science without starting from scratch
42:14
Why geometry might be the missing key to AI-powered physics
48:36

Transcript

Liam Fedus: ultimately, science is 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 exper...