←
Russ Poldrack52 posts
Russ Poldrack avatar

Russ Poldrack

@russpoldrack

Professor at Stanford. Latest book project: Better Code, Better Science (bettercode-book.org/)

◷ Joined October 2009▧ 4 Photos and videos
Russ Poldrack avatar
Russ Poldrack@russpoldrack·Nov 17

This morning I spent time reading code (gasp!) generated by Astra & Fable for our boldtailor project. What I saw was not great: - Despite my strong admonitions to write readable code, it was anything but. In particular, naming of many functions (especially internal ones) was terrible. e.g. a function called _write() that didn't actually write anything. - Unannounced in-place mutation was widespread. In fact, that _write() function actually mutated its input variables in place, with no announcement that it was changing them. - Overuse of custom context managers, sometimes with hidden closeout requirements that were not documented - God-like mutable state object (_State) passed between workflow steps with implicit dependence on earlier steps An AI-generated report describing the issues identified using further AI audits is here: github.com/poldrack/boldtail… This code all worked - the 2571 tests (!) all passed, with 98% code coverage. But we need to expect more of our code than just passing the tests; this package would be a maintenance nightmare. It remains to be seen whether we will be able refactor this into something we will be willing to ship, or instead treat it as a prototype and build a clean version from scratch. I'll keep y'all posted!

Russ Poldrack avatar
Russ Poldrack@russpoldrack·Nov 15

It's been fun to pit Fable 5.2, Astra, and Hydrogen against one another on a complex scientific problem (implementing REML within our boldtailor package). I had Astra create an overview of the development process, which highlights a few things: - today's best models still pass off incorrect code as complete, even when using a test-driven development workflow - adversarial testing across models helps identify errors, but can require multiple rounds - expert judgment in the loop remains important to keep development on track github.com/poldrack/boldtail…

Paul Scotti avatar
⟳ Russ Poldrack retweeted
Paul Scotti@humanscotti·Nov 15

Why do major outlets keep framing brain decoding research as a dystopia in waiting? (NYT this week, MIT Tech Review last month) Maybe it’s the “mind reading” label, or maybe fear just drives engagement, but bookending these pieces with hints that AI will soon read out your deepest secrets, detect your lies, or serve you brain-optimized ads does a disservice to the richer, more optimistic conversation I think the work deserves fMRI in particular is far too noisy and fragile to ever be useful for lie detection or nonconsensual decoding (moving your head a few mm ruins the scan, decoding only works with a cooperative person, MRI is expensive & a huge immovable machine) What we are near are brain foundation models and other neuroAI advances getting good enough to help patients in new ways like reducing symptoms, detecting lesions, tracking clinically meaningful outcomes Where’s the discussion of treating psychiatric disorders with real-time fMRI neurofeedback? Or using brain foundation models to predict disease progression and treatment response, or to speed up drug discovery? These ideas have real research going on right now, not in an unknown dystopian future, and yet with the current media coverage on the research the average person wouldn't know, and may instead be scared rather than optimistic Note: neuroethics & discussion on mental privacy are important! but coverage should spend at least as much time on what this research can do for good as on what could go wrong

fMRI foundation models are cool. Mind-reading hype is not. nytimes.com/2026/11/12/opini…

Post attachment
Fil avatar
⟳ Russ Poldrack retweeted
Fil@BallinFil·Nov 10

Best quote I heard recently If you suck without AI, AI helps you suck at scale

Russ Poldrack avatar
Russ Poldrack@russpoldrack·Nov 14

The No AI Slop manifesto: Outcomes over output Understanding over speed Our own thinking over a model's first draft Better work over easier work Improving the system over patching the output slopmanifesto.com/

Russ Poldrack avatar
Russ Poldrack@russpoldrack·Nov 13

The nice thing about being on Bluesky for the last two years is that I didn't realize there were so many intereseting things I needed to know about. How did I even survive?

Russ Poldrack avatar
Russ Poldrack@russpoldrack·Nov 12

After a couple of years wandering in the wilderness I am back. Just too much interesting stuff happening here.

Russ Poldrack avatar
Russ Poldrack@russpoldrack·Nov 12

Looking forward to this!

🧠 What if we could map the neural basis of human thought—and make neuroscience more trustworthy at the same time? At the Chen Institute 10th Anniversary Symposium, Stanford’s @russpoldrack explores the brain, the mind & the science of reproducibility. 📅 Dec. 3–4, 2026 @Caltech

Post attachment