You can totally imagine the fact that PyTorch ignores the value -100 in loss functions becoming a nerdy plot point in a story about us averting an AI uprising.

🤯 My brain is still reeling a bit:

Accel got *hundreds* of open-source maintainers and founders together in one room this afternoon – the brilliant and hardworking, community-driven contingent behind projects like FastAPI, Ruff and uv, LLVM, Flask, Vue, Vercel, Python and…

Code released for toy N-body simulator: convert any image into N-bodies and let them gravitate

github.com/pererossello/n…

there is a mosquito somewhere in the room. i have 2 hours to find it before it just defeats me

Ten thousand Twitter accounts just sat bolt upright in bed, drenched in cold terror-sweat

Alternate history where Wiener went with the Latin rather than Greek root and called it 'gubernatics' instead of 'cybernetics', leading William Gibson to call it 'guberspace' and 'guberpunk'.

also i do think engagement payouts juice this a bit. everyone being just a little more incentivized to go for that dunk or scathing reply gets amplified by feed dynamics

i understand this is an extremely midwit opinion or whatever but i think you can make compelling arguments in favor of free speech while pointing to clear violations without always trying to maximally embellish the context to fit 10 different objectives

Teknium (e/λ) twitter.com/teortaxesTex/s…

this is definitely the weirdest show i've been a part of and i was in the vibecamp show AND Bet On Love

The algorithmic mirror of a sufficiently powerful recommendation system confidently says, 'based on everything we know about you, this is who you are' and sometimes it's ugly

tired: look how many H100s I have in my datacenter

wired:

debugging the code i wrote about 6 months ago when i was planning to ship and dip

based. Reka continues to impress