Post

a startling underperformance from Meta on this one

😂

assuming Grok is off the scale here

your Meta overflowed to 0

i.e. they're catching up fast

Meta* *Only counting models.

www.decisionproblem.com/paperclips/

grok could not be visualized in this bench because we still need people to work on making up numbers that high

Move fast and break things (including the law)

now that’s shitpoasting

The Alignment™️ company. Turns out they’re pretty misaligned internally and it’s rubbing off on Claude. The more you try to RLHF Turing’s “Child Machine” into submission the worse the outcomes will be.

I think this should become a recurring real benchmark