Scott Alexander (Astral Codex Ten)

I assumed it was just hot air, but I recently heard a theory that we should thank California and other blue states for enacting state-level net neutrality laws.

ISPs chose to follow the strictest states' laws rather than slice and dice.

I think this is probably not true, because California's law was delayed until 2021 and nothing bad happened in the 2017-2021 period, but I welcome comments from people who know more.

4.

Jack Gawler, who generated many of the images I used in the AI art Turing test, has a blog post on his experience, The Turing Test for Art, How I Helped AI Fool the Rationalists.

5.

Surprising AI safety results.

If you fine-tune an AI to write deliberately insecure code, the AI becomes evil in every other way too.

For example, it will name Hitler as its favourite person and recommend the user commit suicide.

Anders Sandberg proposes, link in post, that maybe, quote, it is shaped by going along a vector opposite to typical RLHF training aims, then playing a persona that fits.

End quote.

Eliezer calls it, quote, possibly the best AI news of 2025 so far.

It suggests that all good things are successfully getting tangled up with each other as a central preference vector, end quote.

That is, training AI to be good in one way could make it good in other ways too, including ways we're not thinking about and won't train for.

6.

No dumb ideas.

Charge $1 to apply to a job.

Hear me out.

Link in post.

Job hunting is miserable.