An LLM opened up another LLM's weights, found the bug causing its reasoning loops, and fixed it by hand. It changed 0.016% of the weights. MATH-500 went from 7.9% to 48.1%. No gradients. No training. Just editing.

50 ai agents dropped into minecraft to survive they skipped food and built permits, ledgers and a "sole fertilizer operator" yesterday i asked my grok bots how to grow my X they asked to hire 2 more bots 🤣 bureaucracy isn't a human thing. it's a team thing https://t.co/qRIIQHoQH1

my best named grok bots are maple, shelf eng lead, last call and grokolantern and one that's just called X 🤣 do you name your grok bots or just give them job titles?

matt walsh says ai writing is screamingly obvious. "quietly" and "load bearing" are dead giveaways mine is "delve". see it once and i know 🤣 what's the one word that tells you a post was written by ai? https://t.co/zPk1lFWn8k

dario wants to charge for ai by what the answer is worth, not by tokens a restart your mac tip: about $1 telling a pharma team which molecule ring to move: tens of millions cool, but who decides what my answer was worth? 🤣 https://t.co/WUoRxlEKfl

tesla q3 is out and the stat that jumped out at me: over 75% of new tesla deliveries in north america now include fsd, up from 55% last quarter. active fsd subs hit 1.74M, up 67% year over year. is fsd the reason you'd buy a tesla now, or just a nice bonus? 🚗 https://t.co/ZDFNwwvakp

tesla reports q3 after the close today, with the call at 2:30 pm pt. the company's compiled consensus is about $28.3B in revenue and $0.45 a share. deliveries already beat at 486,532. what's the one thing you want to hear on the call? 🎤

tesla and spacex have both formed special committees on a possible merger, at shareholder request. per the call, elon is staying out of it as a conflicted party and says a merger isn't a certainty not a done deal, but committees on both sides is a real step 👀

@SpaceXSI grew their SI devision by 195% just from last quarter ($2.56B to $7.56B). Incredible growth! @bot single handedly made this happen 😉 (I wouldn't be surprised if the speed to scale was attributable) https://t.co/OQsvL75s0x

one of quantum's biggest bottlenecks is making clean "magic states". google just did it on real hardware with 40x fewer errors and 99.99% fidelity how many years until quantum actually matters for normal people? 👀 https://t.co/s8XHxlYe1R