Ben Recht
@beenwrekt
optimization. machine learning. uc berkeley. I blog at https://argmin.substack.com/ The world won't end.
It's going to be one of those really stupid days on Twitter.
Breaking: 350+ leading AI researchers (including the CEOs of OpenAI, Anthropic, and Google DeepMind) have signed a remarkable statement warning that AI poses a “risk of extinction,” and comparing it to pandemics and nuclear weapons. nytimes.com/2023/05/30/tec…
"Prediction markets are a core part of news consumption." It's true, and also bad.
Two media takeaways from last night's election: - Prediction markets are a core part of news consumption. The bettors aren't always perfect, but they're part of how we follow developing events now. - NY1 is such a good network if you're going to watch TV for local politics.
Very half-baked philosophy of engineering post: How do we prove that something is unpredictable in machine learning? argmin.net/p/the-unpredic…
Responding to @JessicaHullman, I propose an alternative to checklists and statistical mandates in machine learning research. argmin.net/p/metascience-…
About that AI coding study and why quantitative “measurements” in social science are always ethnographic vibes... argmin.net/p/are-develope…
Fully open machine learning requires not only GPU access but a community commitment to openness. (Some nostalgic lessons from the ImageNet decade.) argmin.net/p/an-open-mind…
Testing the waters of finance blogging, I wrote about deep learning memes, random features, and the glory of kernel smoothing. argmin.net/p/you-keep-usi…
The NeurIPS paper checklist corroborates the bureaucratic theory of statistics. argmin.net/p/standard-err…
To figure out the purpose of forecasting, I put on my Dan Davies hat and ask, “What do forecasts do?” argmin.net/p/one-out-of-f…
Forecasts as probabilities of the ruthless Bayesian. argmin.net/p/probability-…
I’m declaring it forecasting week on the blog. Here’s a cute example transmuting errors of averages into averages of errors. argmin.net/p/restatements…
In a new paper, Juanky Perdomo and I ask when forecasting and prediction can be solved by sneaky accounting. The answer turns out to be “more often than you’d expect." argmin.net/p/in-defense-o…
Validity as a style guide for science. argmin.net/p/strunk-and-w…
p-values supposedly measure the correlation between interventions and outcomes. But what happens when a measure becomes a target? argmin.net/p/milton-fried…
I use the vetting of papers as a case study for how academic process evolves. argmin.net/p/the-open-mar…
I wrote a defense of peer review, as it will be the system academia uses to reinvent itself. argmin.net/p/a-defense-of…
Who should get to define which science is "good" and "bad"? argmin.net/p/the-good-the…
I weigh in on the Trump administration’s newfound obsession with Gold Standard Science and reproducibility. Though it’s not all in bad faith, it’s likely to backfire. argmin.net/p/this-is-fine
Louis Fein’s utopian vision of academic computer science was delightfully boring. argmin.net/p/may-you-live…
George Forsythe's 1962 vision for computer science education. argmin.net/p/physics-for-…
On the cultivation of a "pure" computer science and why we should reject this concept. argmin.net/p/computationa…
