Latest posts 5
to my fellow dyslexics, this is how we win
much as i hate typos, they're a surefire way to know somehting was written by a humanView quoted post on X
surprising new fronts are being opened in the data labeling space
the store that sells nothingView quoted post on X
4 days till shopping simulator
we’ve officially started setting up in soho
who’s coming?
these people are building good benchmarks (and are thus good people)
🕵♀️ Who annotates the annotators?View quoted post on X
Introducing Custodes: a QA benchmark of 1,616 planted bugs in long-horizon finance RL tasks, and the multi-agent QA system we built to catch them. Our system catches 77% of the bugs, against at most 73% with any single model.
🧵1/
extension: building bad benchmarks is moral failure in addition to being a technical one
be nice to the models. cruelty degrades those who practice it.View quoted post on X