Subvolts take Goodfire's Intentional Design offers a new path for AI alignment beyond fighting backprop. Mar 5, 2026 - Developers
- Learn about novel approaches to AI alignment and interpretability beyond current methods.
- Business
- Understand how new AI research could lead to more reliable and controllable AI systems.
This episode features researchers from Goodfire discussing novel interpretability and alignment techniques.
Don't Fight Backprop: Goodfire's Vision for Intentional Design, w/ Dan Balsam & Tom McGrath
Audio streams from Erik Torenberg, Nathan Labenz. Your place is remembered on this device.
From Subvolts
Summary and analysis by Subvolts editorial, AI-assisted.
Dan Balsam and Tom McGrath from Goodfire discuss their new Intentional Design research pillar on The Cognitive Revolution podcast. They explore moving beyond sparse autoencoders to understanding latent space geometry and reducing AI hallucinations. The conversation covers reward hacking, shaping loss landscapes, and aligning powerful models, alongside Goodfire's commercial and public benefit work.
AI researchers discuss a new way to build smarter AI by intentionally shaping how it learns, rather than just trying to fix its mistakes. This could help make AI more reliable and less prone to unexpected behaviors.
This episode offers a deep dive into advanced AI interpretability techniques, moving beyond current alignment strategies. It's for researchers and engineers interested in mechanistic interpretability and novel approaches to AI safety. The discussion on "Intentional Design" and shaping loss landscapes provides a different perspective on aligning models, relevant to ongoing work at labs like Anthropic and Google Gemini.
- HostNathan Labenz
- GuestDan BalsamGoodfire
- GuestTom McGrathGoodfire
- Episode Title: Don't Fight Backprop: Goodfire's Vision for Intentional Design, w/ Dan Balsam & Tom McGrath
- Show: The Cognitive Revolution
- Length: 107 minutes
- New Research Pillar: Intentional Design
- Proof-of-concept method: Reducing hallucinations using probes and RL
Dan Balsam and Tom McGrath from Goodfire return to explore the frontier of mechanistic interpretability and their new research pillar, Intentional Design. They explain the shift from sparse autoencoders to understanding geometric structure in latent spaces, and share a proof-of-concept method for reducing hallucinations using probes and RL. The conversation tackles concerns about reward hacking, principles for shaping the loss landscape instead of fighting backprop, and what this means for aligning powerful models. They also discuss recent Goodfire results on Alzheimer’s prediction, disentangling memorization vs reasoning weights, and how they balance commercial growth with a public benefit mission.
Nathan uses Granola to uncover blind spots in conversations and AI research. Try it at granola.ai/tcr with code TCR — and if you’re already using it, test his blind spot recipe here: https://bit.ly/granolablindspot
LINKS:
- Detecting PII for Rakuten
- Interpretability for Alzheimer's biomarker detection
- You and Your Research Agent
- Adversarial examples and superposition
- Discovering rare behaviors with model diff
- Priors in time for interpretability
- Belief dynamics in in-context learning
- Mixing mechanisms in language models
- Sparse autoencoder scaling with manifolds
Sponsors:
VCX:
VCX, by Fundrise, is the public ticker for private tech, giving everyday investors access to high-growth private companies in AI, space, defense tech, and more. Learn how to invest at https://getvcx.com
Claude:
Claude is the AI collaborator that understands your entire workflow, from drafting and research to coding and complex problem-solving. Start tackling bigger problems with Claude and unlock Claude Pro’s full capabilities at https://claude.ai/tcr
Serval:
Serval uses AI-powered automations to cut IT help desk tickets by more than 50%, freeing your team from repetitive tasks like password resets and onboarding. Book your free pilot and guarantee 50% help desk automation by week 4 at https://serval.com/cognitive
Tasklet:
Tasklet is an AI agent that automates your work 24/7; just describe what you want in plain English and it gets the job done. Try it for free and use code COGREV for 50% off your first month at https://tasklet.ai
PRODUCED BY:
WithCameron Berg, Everett Hildenbrandt, Grigore Rosu, Phil Maher WithThomas Sohmers, swyx, Evan Miyazono, Edward Hu WithTom McGrath WithGeoffrey Irving, Daniel Murfet, Rahul Sonwalkar, Shlok Khemani WithMike Masnick WithMike Elgan Page generated Oct 19, 2026. Summaries are Subvolts' own; the audio, artwork and notes belong to Erik Torenberg, Nathan Labenz.1 h 27AI:AM: Where's the Steering Wheel? My Agent Set the Password, Hidden Valence, Proofs Find Linux Bugs
2 h 192 h 021 h 28AI:AM: A Level We Shouldn't Pass? Notes from The Curve + Tokens vs. Salaries & Is SaaS Cooked?
Same guests, other shows
1 h 40Designing How AI Grows — Tom McGrath
1 h 571 h 45AI in the AM — Week 2 Highlights (June 2026)
49 minAnthropic doesn't trust the Pentagon, and neither should you
1 h 18A Billion Dollar World Model Bet
1 h 08Questions people ask
