Latent Space: The AI Engineer Podcast

Episode: 🔬Searching the Space of All Possible Materials — Prof. Max Welling, CuspAI

Date: February 25, 2026

Overview

This episode of Latent Space dives into the revolution of "AI for Science," centering on the intersection of advanced machine learning and material science. Renowned AI researcher and CuspAI co-founder, Prof. Max Welling, joins the discussion to explore how generative AI and automated experimentation are radically accelerating the search for new materials—pushing us toward solutions for climate change and unlocking transformative industrial possibilities. The episode covers the conceptual threads of physics in Max’s work, practical challenges in building AI platforms for materials discovery, and the philosophical and technical nuances of merging science and engineering at this new frontier.

Key Discussion Points & Insights

1. Physics as the Common Thread in Welling’s Career

• Physics as the Unifying Theme: Max explains that his work—from quantum gravity to variational autoencoders and graph neural networks—has always been guided by a deep fascination with physics and the mathematical tools it provides.
  • “Physics is the thread. So having done, you know, spent a lot of time in theoretical physics, I think there is first very fundamental and exciting questions, like things that haven't actually been figured out in quantum gravity…”
    — Max Welling, 03:41
• Symmetry in ML: He describes how concepts like symmetry, core in physics (e.g., gauge symmetry), have shaped cutting-edge machine learning models.
• Impact Over Curiosity: Initially driven by pure curiosity, Welling now balances interesting science with the desire to make real-world impact, particularly regarding climate change.
  • “So as I get closer to retirement…I do want to kind of make a positive impact in the world. And I got pretty worried about climate change.”
    — Max Welling, 02:33

2. The Rise of “AI for Science” and Materials Discovery

• Why Now? Developments like AlphaFold (protein folding prediction) and machine-learned force fields have demonstrated AI’s utility in science, spurring enormous investment and excitement in the sector.
  • “It's exploding...now actually a startup by Jeff Bezos...6.2 billion seed round. Right. It's like insane.”
    — Max Welling, 07:19
• Materials as the Foundation: Welling emphasizes that beneath every advanced technology is a materials challenge. Whether it’s GPUs for AI or batteries for energy transition, innovation is now bottlenecked by the speed of discovering new, better materials.
  • “Underlying almost everything is a material...the very foundation of what you're doing is a material problem.”
    — Max Welling, 11:30

3. CuspAI: Vision and Platform Approach

• Startup Motivation & Scale: CuspAI emerged from a desire to confront climate change by enabling rapid and automated discovery of sustainable materials.
  • “We realized ... to stay within 2 degrees, let's say we would not only have to reduce our emissions to zero by 2050, but then ... another half century or even a century of removing carbon dioxide from the atmosphere ... That is an unsolved problem.”
    — Max Welling, 14:30
• Digital and Physical Automation: The platform combines generative AI for proposing new candidate materials, a “digital twin” stack for filtering and simulating candidates, and high-throughput experimental automation tying back real-world results (“physics processing unit” or “PPU”).
  • “It's basically nature doing computations for you. It's the fastest computer known possible. Even it's a bit hard to program because you have to do all these experiments.”
    — Max Welling, 00:00 / 14:30
• Practical Evolution: The system is both modular and iterative—starting with lots of human-in-the-loop workflows and progressively automating components as confidence and toolkits improve.
  • “You build all these tools and then you go through a workflow ... just manually. And then you build the agent...”
    — Max Welling, 21:00
• Human Expertise Remains Crucial: Fully automated “dark labs” are not the goal. Instead, AI is a super-powered assistant for chemists and material scientists, helping them work faster and smarter.
  • “We don't see something very soon where the chemist and domain expert is out of the loop ... it's an increasingly powerful tool in the hands of the chemist.”
    — Max Welling, 23:05

4. Incremental Progress vs. Breakthroughs in Materials

• Mixed Strategy: CuspAI pursues both “lighthouse” moonshot materials and smaller, incremental industry projects—believing both drive value and reinforce each other.
  • “We follow a mixed strategy, so we are definitely going after a big material ... as a proof point. At the same time, we also are quite happy to work with companies that have more modest goals.”
    — Max Welling, 24:46
• Immediate Usefulness: Unlike other hard tech fields, new tools and models in AI for materials are useful immediately, even before the proverbial “big win.”
  • “Every time you build something, it's actually immediately useful. Right. And so unlike quantum computing, which. Or nuclear fusion ... here, every time you introduce, so you go to a customer and you say, so what do you need?”
    — Max Welling, 26:04

5. Technical Deep Dive: Symmetry, Equivariance, and Model Building

• Equivariance Defined: Welling provides an accessible explanation—equivariance allows a model to “hard code” symmetry (e.g., a bottle is still a bottle if rotated), which can dramatically reduce required training data.
  • “Equivariance is the infusion of symmetry in neural networks ... once you've trained it in one orientation, it will understand it in any other orientation.”
    — Max Welling, 28:44
• Data Augmentation vs. Equivariance: Sometimes hardcoding symmetry is better, sometimes classic data augmentation wins—the field observes a delicate optimization tradeoff.
• Inductive Bias vs. Scale: The “bitter lesson” is about balancing the incorporation of mathematical prior knowledge (inductive bias) with scalability and the power of large datasets.
  • “It's a trade off between data and inductive bias. So if your inductive bias is not perfectly correct, you have to be careful because you put a ceiling to what you can do.”
    — Max Welling, 30:50

6. Making AI for Science Accessible & How to Get Involved

• Bridging Background Gaps: Max advocates for more cross-disciplinary training and open educational resources (books, courses, workshops). He’s actively contributing with a new book to help engineers dive in.
  • “We should create curricula that are on this interface ... there is a huge amount of content you can, you can go and see.”
    — Max Welling, 10:14

7. Upcoming Book: Connecting Generative AI and Physics

• Key Message: Welling’s new book, Generative AI and Stochastic Thermodynamics, draws mathematical parallels between AI diffusion models and nonequilibrium physics, enabling richer cross-pollination of ideas.
  • “It basically lays bare the fact that the mathematics that goes into both generative AI ... and this field of non equilibrium statistical mechanics ... is actually identical.”
    — Max Welling, 31:37
• Potential Impact: The hope is that unifying perspectives will “make our algorithms better” and help scientists pursue new lines of research.

Notable Quotes & Memorable Moments

• On Physics as Computation:
  “It's basically nature doing computations for you. It's the fastest computer known possible.”
  — Max Welling, 00:00
• On Motivation:
  “As I get closer to retirement ... I do want to kind of make a positive impact in the world. And I got pretty worried about climate change.”
  — Max Welling, 02:33
• On the Role of Materials:
  “Wherever you go ... actually the very foundation of what you're doing is a material problem.”
  — Max Welling, 13:00
• On Platform Philosophy:
  “For me, it's really empowering the domain experts that are sitting in the companies and in the universities to be much faster in developing their materials.”
  — Max Welling, 23:05
• On AI/Physics Cross-Fertilization:
  “When we see that these things are actually the same, we can ... look at this new theory that's out there ... and say, okay, what can we take from here that will make our algorithms better?”
  — Max Welling, 33:00

Timestamps for Key Segments

| Timestamp | Content | |------------|---------------------------------------------------------------| | 00:00 | Physics as computation; "physics processing unit" analogy | | 01:35 | Early motivation and evolution of impact-driven research | | 03:41 | Physics as the common thread; symmetries and deep learning | | 07:06 | The explosion of "AI for Science" and industry investment | | 09:43 | Getting involved in AI for science as a non-scientist | | 11:30 | Why materials are foundational for AI engineers and beyond | | 14:30 | CuspAI origin story, growth, and model for material search | | 17:32 | How the CuspAI platform works; integration of AI and nature | | 20:56 | Automation vs. human-in-the-loop and platform evolution | | 24:46 | Lighthouse projects vs. incremental industry partnerships | | 26:04 | Why progress in this field is immediately useful | | 28:44 | Explanation of equivariance and symmetries in ML | | 31:37 | Upcoming book: Generative AI and Stochastic Thermodynamics |

Concluding Thoughts

The episode provides both a high-level vision and practical blueprint for how AI is set to transform scientific discovery—starting with materials, but openly gesturing at a future where algorithmic and physical progress mutually accelerate. Prof. Max Welling’s passion for physics, humility about the field’s challenges, and optimism about collaborative intelligence (human + machine) offer a hopeful, grounded picture of the coming decade in AI for science.

For more information or show notes, visit latent.space.