What if neural networks are less like mysterious black boxes and more like systems we can inspect, debug, and eventually design with intention?

In this episode of The Neuron: AI Explained, Corey Noles and Grant Harvey talk with Eric Ho, Cofounder & CEO of Goodfire, an AI interpretability company working to understand what’s happening inside neural networks. Eric explains why models may contain meaningful internal structures — including features, representations, circuits, and curved manifolds — and how mapping those structures could make AI systems safer, more reliable, and more useful.

They discuss why models may “think in shapes,” how Goodfire uses AI to interpret other AI systems, what neural geometry can reveal about hallucinations and model behavior, and why interpretability could change how companies train and control their own models.

They also get into consciousness, robotics, multimodal models, the bitter lesson, and why Eric thinks more people should be looking under the hood of the most consequential technology of our time.

Subscribe to The Neuron for more grounded conversations about how AI actually works: https://www.theneuron.ai/

Sponsored by SAS AI Governance: Visit https://www.sas.com/en_us/solutions/ai/governance.html?utm_source=other&utm_medium=cpm&utm_campaign=sas-vpl-gbc-gps-global#

The Neuron Academy helps professionals build practical AI skills they can use right away, with lessons on prompting, workflows, and real workplace use cases. Check out https://theneuronacademy.com/ today!