i now have an ideas/questions/hare-brained schemes slack channel — if we've spoken, feel free to DM if you'd like to be added :)

one of my most strongly held and least consequential beliefs is that knightian uncertainty just isn't a real thing

If a model card reports the results of dangerous capabilities evals and doesn't specify that they were conducted with safeguards removed, you should assume that safeguards were in place for testing.

Short 🧵 👇

so who is making huelnog for this festive season ?

i think RAND's report on securing ai model weights is great, i also think that if you read the whole thing you should get a little sticker or a pin badge or something

i want to like bluesky but it doesn't quite have that disgusting eating pennies/cutting your tongue/looking too long at the sun feel that the other place has