thanks Arvind Narayanan and Matthew Salganik for the support!

truly excited to be starting my cs phd at princeton this fall 🥳

The way a lot of ppl seem to see the world is:

- if things are getting bad, that's news
- if things have been bad but are getting better, that's also bad news
- if things are good, that's boring so it's not news

The problem with this approach isn't just that you persuade…

One of the greatest MYTHS in cultural psychology is collectivists (e.g. Asians) don't compete––they are a kumbaya/harmonious/lovey dovey bunch who prioritize interpersonal relationships, unlike those individualistic Americans who are self-serving and competitive. 1

Current benchmarking practices measure only capability, not reliability. The variance matters as much as the mean. If a model scores 80%, does it mean it works 80% of the time on each instance or 100% of the time on a subset? Totally different implications for downstream users.

This is the craziest example of the ongoing AI frenzy I've ever seen. As far as I can tell this means that a single developer can make queries that burn $1,400 worth of compute for free *per day* (e.g. 'reverse the following 1M token string') — 50 * (7+21) = 1,400.
HT Simon Willison

Back in November, Google announced 2.2 milllion new materials.

Today, a paper in Chemistry of Materials from Ram Seshadri and Tony Cheetham dismantles that claim

pubs.acs.org/doi/10.1021/ac…

I'm ecstatic to share that preorders are now open for the AI Snake Oil book! The book will be released on September 24, 2024.

Arvind Narayanan and I have been working on this for the past two years, and we can't wait to share it with the world.

Preorder: princeton.press/awecv19z

Do Community Notes reduce the spread of false information? Short answer: Yes. Longer answer: Yes, but... A 🧵  arxiv.org/abs/2404.02803

There's nothing exceptional about tech policy that makes it harder than any other type of policy requiring deep expertise. If we can do health or nuclear policy, we can do tech policy.

Plus new writings and an upcoming event for policymakers aisnakeoil.com/p/tech-policy-…
w/ Sayash Kapoor

Professor Anderson was a giant of the field. Yet he often took the time to drop me a note of appreciation when he liked something I published. It meant a lot to me, especially as a junior scholar. I corresponded with him just last week. This is a complete shock. RIP Ross

Some brutal victim statements from FTX customers
courtlistener.com/docket/6663129…

Responses to NTIA on open FMs

1. On marginal risk
hai.stanford.edu/sites/default/…
Led w/ Sayash Kapoor Kevin Klyman Dan Ho Percy Liang Arvind Narayanan

2. On release strategies, thresholds
ias.edu/sites/default/…
Led by Alondra Nelson Christine Custis Dorothy Chou Marc Aidinoff Séb Krier William Isaac

Would Boeing murder a whistleblower? Before this month I'd have said it depends. I didn't think Boeing would kill a middle manager who never worked on the 737 Max & left in 2017.

What surprised me was how many more informed observers felt Boeing...might
prospect.org/infrastructure…

Been saying all this for a while!
aisnakeoil.com/p/ai-safety-is… AI safety is not a model property
aisnakeoil.com/p/model-alignm… Model alignment protects against accidental harms, not intentional ones
aisnakeoil.com/p/licensing-is… Licensing is neither feasible nor effective for addressing AI risks