This new center strikes the right tone in approaching the AI alignment problem. alignmentalignment.ai

On unquantifiable costs and inherent tradeoffs in decision theory. www.argmin.net/p/stuck-in-t...

How Snake Oil Becomes Normal Technology: A Brief History of American Capitalism

This was a catty subtweet, but I swear I am earnestly interested in an answer. There are a few good ones in the replies.

Does anyone have recommendations for studies or books on how snake oil becomes normal technology?

Three different views of optimal decision theory yield identical decisions. Which view is right?

In a new paper, I try to resolve the counterintuitive evidence of Meehl’s “clinical vs statistical prediction” problems: Statistics only wins because the game is rigged.

Lecture 2 talks about actuarial prediction, why it’s easy, and how a statistical view declares all variation is “noise.”

How we’ll be evaluated in the future determines our machine learning methods in the present.

One extra AI Safety tidbit for this morning: Anthropic's Claude refused to transcribe this passage by Allen Wood (h/t @hilzoy.bsky.social) because it violated their safety protocols.

I wrote about the tragic death of Adam Raine and the venal negligence of "AI Safety." www.argmin.net/p/the-banal-...

As promised, a syllabus for this semester’s graduate course on machine learning.

Optimization folks, it turns out that BFGS is sentient.

Should we still teach machine learning classes? Let me justify my existence.

Are linear models simple and interpretable? My answer won’t surprise you.

I'm really fond of John Ganz's coinage of "Vulgar Positivism." The Argument thinks it can make a profit embracing maximum vulgarity.

The new magazine promising to "publish people who seek truth from facts" opens with an overconfident analysis of cash transfer RCTs. Cursed timeline.

Should machine learning care about simple baselines?

This might be the greatest random YouTube recommendation I've ever received. www.youtube.com/watch?v=vaG5...

If machine learning is a game, it’s Calvinball. Bitter lessons from chess don’t apply.

Since it's 5 PM here, let me follow up this post with my top 10 negroni variations.

Revisiting Sutton’s Bitter Lesson in the wake of GPT-5.

Everyone who uses the term "inductive bias" means something different.

The American Analog Set/Death Cab for Cutie crossover experience? I'll allow it.

I've started thinking about my fall semester syllabus.

The version on the METR website has some helpful labels on the y-axis to give you a feel for how silly this plot is. metr.github.io/autonomy-eva...

Hey quant social science folks: how do you think OpenAI measured the points in this graph? (My guess: they asked ChatGPT to fake a plot that looked impressive. It's good at that sort of thing.)

live by the statistical sword, die by the statistical sword.

I'm hearing OpenAI's new oss model is ass.

I hereby welcome John Ganz to the Resistance.

Fun fact: this Rationalist temple is a block from my house. Also a few steps away from the Hari Krishnas. AMA.

I’m excited to announce that my new book, _The Irrational Decision_, is now available for pre-order from Princeton University Press.

"It's called transfer learning if it works and overfitting if it doesn't."

These three things seem deeply interconnected: 1) app gambling culture. 2) the only industry adding jobs is health care. 3) AI cap ex now more than all consumer spending combined. I need to write a ted talk about how.

The American economy now stands upon the abandonment of clarity and efficiency in software engineering.

Very half-baked philosophy of engineering post: How do we prove that something is unpredictable in machine learning?

Responding to @jessicahullman.bsky.social, I propose an alternative to checklists and statistical mandates in machine learning research.

Based on a fun discussion here last night, I wrote about that AI coding study and why quantitative “measurements” in social science are always ethnographic vibes.

Stats friends... what would your estimator be if you were interested in a similar question as this study that is lighting Bluesky on fire tonight? 1/x

Fully open machine learning requires not only GPU access but a community commitment to openness. (Some nostalgic lessons from the ImageNet decade.)

Testing the waters of finance blogging, I wrote about deep learning memes, random features, and the glory of kernel smoothing.

This is a question that deserves a blog post answer, but I'm going to do a quick response here. Gather around children as I tell you a story from ancient times. The year was 2015... at that point, it appeared that you needed Google scale to build competitive computer vision systems. 1/x

I strongly endorse the aspirations of this post by @natolambert.bsky.social. I am curious: why *is* there no American Deepseek? I don't think it's solely because of a lack of effort.

The NeurIPS paper checklist corroborates the bureaucratic theory of statistics.

Do you remember NeuRIPS 2013, when Zuck held court in the penthouse of the Lake Tahoe Harrah's, desperately trying to woo a select group of attendees to join FAIR?

I'm sorry, there's a list? And those are three of the names on it? I'm going to go stick my head in the sand now.

Language Machines by @leifw.bsky.social is out! Highly recommend this book for a thoughtful, humanistic take on what LLMs reveal about language itself.

The polling did not prove accurate. I wonder what our friend here learned.

how about them polls?

This ad outside the Equinox in Vancouver is haunting me.