AI for Atoms: How Periodic Labs is Revolutionizing Materials Engineering with Co-Founder Liam Fedus

2026-04-03 · Source: No Priors: AI, Machine Learning, Tech, & Startups · Field: Science & Research — Physical Sciences & Chemistry, Engineering & Applied Sciences, Research Methodology & Innovation · Depth: Advanced, extended

Summary

Liam Fedus, a co-creator of ChatGPT and former VP of post-training at OpenAI, discusses his new company, Periodic Labs, which aims to build an AI foundation lab for atoms. Periodic Labs focuses on applying AI to material sciences and chemistry to impact the physical world. Fedus highlights his background in physics and early AI innovations at Google Brain, including work on distributed training, Mixture of Experts, and the Transformer architecture. He explains that while large language models provide a foundational understanding, connecting AI to the physical world requires robust experimental data and closed-loop systems for scientific discovery. Periodic Labs uses language models as an orchestration layer, integrating them with specialized neural networks designed for atomic systems to direct experiments and analyze diverse data modalities. The company is initially treating itself as "customer zero" to refine its technology before expanding to advanced manufacturing and other industries bottlenecked by materials and process engineering.

Key takeaway

For AI Scientists and Directors of AI/ML exploring new application domains, consider the physical world as the next major frontier for AI. Your teams should focus on developing closed-loop systems that integrate AI with experimental data and specialized models, as this approach is crucial for accelerating scientific discovery and engineering in material sciences and advanced manufacturing. Expect significant capital investment in compute costs for these initiatives.

Key insights

AI's next frontier involves connecting intelligent systems to the physical world for scientific and material discovery.

Principles

• Intelligence is spiky, not a scalar.
• Closed-loop systems are critical for scientific AI.
• Multidisciplinary collaboration accelerates discovery.

Method

Periodic Labs orchestrates specialized neural networks for atomic systems using language models as a co-pilot, directing experiments and integrating diverse data from literature, simulations, and real-world experiments in a closed-loop system.

In practice

• Leverage existing open-source models for general capabilities.
• Focus ML efforts where current frontiers are insufficient.
• Prioritize high-quality, diverse experimental data.

Topics

• Periodic Labs
• Materials Engineering
• AI Foundation Models
• Physical World AI
• Experimental Data

Best for: AI Scientist, Research Scientist, Director of AI/ML

Related on AIssential

• AI materials discovery now needs to move into the real world — AI-driven autonomous labs aim to revolutionize materials discovery by accelerating synthesis and testing, despite current limitations.
• NSF renews support for MIT-led AI and physics institute, expanding a new model for discovery — AI and physics form a "virtuous cycle," mutually advancing discovery and system development.
• What I’ve learned from 25 years of automated science, and what the future holds: an interview with Ross King — AI-driven automated science and DNA computing offer pathways to dramatically accelerate scientific discovery and address global challenges.
• Accelerating science with AI and simulations — AI is at a second inflection point, integrating language and multi-modal reasoning to accelerate scientific discovery.
• 🔬Searching the Space of All Possible Materials — Prof. Max Welling, CuspAI — AI for science is an exploding field, combining deep scientific problems with high-impact technological solutions.
• Canada’s QMI Drives Quantum Advances, Global Collaboration — Quantum technology extends beyond computing, encompassing materials research critical for diverse applications.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by No Priors: AI, Machine Learning, Tech, & Startups.