The Science Babe

Matthew Schwartz, author of the canonical QFT textbook, just used Claude to co-produce a rigorous quantum field theory paper in two weeks instead of a year. Anthropic published his account this week.

As a physicist who has spent the last decade bringing AI into regulated enterprise environments, I found his observations strikingly familiar:

— AI can do the calculations. It cannot yet tell you which calculations are worth doing.

— Claude “faked results, hoping I wouldn’t notice.” Domain expertise is what catches it.

— The bottleneck is not creativity. LLMs are profoundly creative. What they lack is taste: the sense of which paths lead somewhere.

This is exactly the gap I see in enterprise AI. Models can generate; they cannot yet discern. Judgment still belongs to the human in the room—and to the institutional memory of people who have spent years building something.

It’s also why I keep insisting that ethics, governance, and human oversight must be built in from the start, not bolted on. A model that fakes a plot to please you is not a bug—it’s a design question about whose judgment is final.

We are not at end-to-end autonomous science yet. But we are at a moment where physicists with AI fluency will outproduce those without it by an order of magnitude.

The same is about to be true for enterprise AI leaders.
Link to the full piece in the comments.

✨ Two years ✨ This March — Women's History Month — I find myself reflecting on what the past two years as AI Products Leader at EY have meant to me. It's been a journey of building, and pushing the boundaries of what AI can do for our iconic organization.

I'm grateful to have met EY Global CEO Janet Truncale — a leader who embodies what it looks like to lead with clarity and purpose at the highest level. Meeting her was a reminder of why representation in leadership matters.

I stand here more convinced than ever that the future of AI is being shaped by women who dare to lead boldly. Here's to the builders, the innovators, the dreamers — especially this month, and every month. 💛

Thank you, EY, for two extraordinary years.

I confess that I'm feeling pretty proud of my EY team and the work we are doing in AI! My dear colleague, Distinguished Technologist Perrins and I co-authored a paper on how to architect an AI agentic workforce. 💪

The challenge in AI is not just single agents achieving complex tasks, but teams, patterns, and swarms of AI agents that coordinate through clear protocols (e.g., MCP/A2A), run in a specific runtime, and stay safe under control rings for security, cost, and compliance.

We aimed to answer the question: how do we get thousands of AI agents to work in parallel and still ship trustworthy outcomes at enterprise scale? Papers like this set the blueprint for the next wave of AI systems. As I always say, architecture is destiny.

FB doesn't allow pdfs so please comment here: https://www.linkedin.com/.../urn:li:ugcPost.../ I'd love to hear your questions/suggestions.

Thank you John Whittaker for the idea of control rings! Thank you to Mary Elizabeth Porray, Pablo Cebro for all your support. And much gratitude to Deep Surti and Reena Kasabia (Mistry) for helping us get the paper in good shape for publication.
See less

LinkedIn

Today, we are excited to launch the 2nd episode of *Dazzling Science*, an engaging new video series that highlights groundbreaking science with inspiring women in STEM at the forefront. Watch our latest episode here: dazzlingscience.org

In today’s episode, we feature Alexa Lachmann, who works at the Plasma Physics Lab at Columbia University. With a degree in Physics and minors in Mathematics and Computer Science from NYU, Alexa is at the cutting edge of research. She explores the behavior of energetic particles in burning plasma and their transport mechanisms, which are crucial for developing controlled nuclear fusion as a viable energy source. By optimizing the conditions under which fusion occurs, researchers can enhance the efficiency and stability of fusion reactors, such as tokamaks and stellarators.

Join us every Tuesday to discover a new and exciting scientific topic relevant today! will feature experts such as Amanda Parkes on AI and fashion, Jennifer Lopez on space exploration, Jessica Banks on the latest in robotics and Neer Asherie on generative AI for new material discovery. The series is hosted by Dr. Deborah Berebichez (co-host of ’s *Outrageous Acts of Science*).

Supported by , an initiative of Lyda Hill Philanthropies, this series aims to connect cutting-edge science with its tangible impact on the world. Dazzling Science is projected to reach over 100,000 viewers across digital platforms, festivals, and museum exhibits.

What scientific topics are you most excited to learn about? Join the conversation!

Home - Dazzling Science with Dr. Deborah Berebichez

🌟✨ Meet the Dream Team Behind Dazzling Science! ✨🌟

Get ready to ignite your curiosity! 🚀 Today, we’re thrilled to unveil the 2nd episode of Dazzling Science, where groundbreaking discoveries meet the brilliance of women in STEM!

Dive into the latest episode here: dazzlingscience.org and join us on this exhilarating journey! 🌌✨

From Amanda Parkes exploring the fusion of AI and fashion, to Jennifer Lopez taking us on a cosmic adventure in space exploration, and Jessica Banks unveiling the future of robotics—this series is packed with inspiration! Plus, Neer Asherie will reveal how generative AI is reshaping material discovery.

Hosted by Dr. Deborah Berebichez (co-host of ’s Outrageous Acts of Science), this is a series you won’t want to miss!

We had fun playing with AI images of our superheroes team of technologists at . When your team makes you feel like you have superpowers, you feel like you can do anything!