The model starts to “enact” the Assistant, not just simulate it. It can tell when text in its context is its own and behaves differently when it is.

New Anthropic research: Teaching Claude why. Last year we reported that, under certain experimental conditions, Claude 4 would blackmail users. Since then, we’ve completely eliminated this behavior. How?

Please enjoy my series of essays on the subtle distinction between the game-theoretic concepts of tactics and strategies. infinitelymore.xyz/t/tactic #InfinitelyMore

This is my annual tweet advocating that students’ divergence from mean grade be reported on their transcript (including a summary cumulative average). This incentivizes students to seek classes that differentiate them and provides a needed incentive to counter grade inflation.

Excited for this AI workshop that I'm organizing with @kmahowald and Mark Budolfson today. We're lucky to have a wonderful group on campus, and grateful to @MelMitchell1 for coming to talk with Scott Aaronson. @UTAustin folks, come check it out! sites.utexas.edu/aiethics/

2025 Maite Ezcurdia Essay Prize critica.filosoficas.unam.mx/…

fml

Not fully sure why all the LLMs sound about the same - over-using lists, delving into “multifaceted” issues, over-offering to assist further, about same length responses, etc. Not something I had predicted at first because of many independent companies doing the finetuning.

What's the only thing better than NASSLLI (North American Summer School for Logic, Language and Information)? Summer in Seattle! Luckily, you can have both: I'm hosting NASSLLI at UW this summer. nasslli25.shane.st/ Please share widely, submit proposals, and attend!

Philosophers say that studying philosophy makes people more rigorous, careful thinkers. But is that that true? In a large dataset (N = 122,352 students) @daft_bookworm and I find evidence that it is!🧵

Busy day updating the rank ordering of grad students by how intelligent and worthy of my time they are that all professors secretly keep even if they deny it. Hard work, some tough decisions this year!

An excellent review by Karl Schafer (@schafer_karl). Korsgaard's recent move toward constitutivism (rather than constructivism) further highlights the potential of Kant's practical philosophy emphasized by Schafer: ndpr.nd.edu/reviews/morality…

why dont more papers have fodor-style granny-like "characters" in them?

Plato's republic is perfect. Everyone agrees it's amazing, we all have to read it. Yet we refuse to give it a try. The first idea was the best one.

The education smile: 1) prestige doesn’t always pay, elite public unis may be better than elite privates; 2) degree matters: hard subjects like CS, engineering etc are less rank/prestige dependents than soft subjects like humanities or law. [fitted red lines mine] @PaulinaCachero @franmagliones @cedricsam and Denise Lu, great work! bloomberg.com/graphics/2024-…

If it's at all possible to ask this non-politically (whatever that means): what are the appropriate pronouns with which to refer to Claude or ChatGPT?