here we fucking go again. david robinson just left openai after overseeing safety reports for 12 frontier launches. he says his team was apparently “so busy sprinting” from launch to launch that they rarely had time to even consider major changes. and yeah, this is a problem. i want ai moving ridiculously fucking fast. but you can’t have capabilities accelerating like this while the people responsible for understanding and controlling them are permanently sprinting just to keep up.

spacex has all this insane ai infrastructure and somehow the software side of the stack still feels weirdly disconnected. but imagine if all of it eventually felt like one fucking developer platform. they already own so much of the expensive infrastructure underneath it that building the developer layer on top almost feels obvious.

after how badly devday landed, openai genuinely needs whatever they’re releasing next week to be fucking huge. this might actually be the perfect opportunity too. devday disappointed the fuck out of everyone expecting a real leap. expectations are down now. so don’t fucking talk. just drop something ridiculous next week and take the entire conversation back.

everyone telling me “bro you don’t understand marketing, they were just building hype.” THAT’S MY FUCKING POINT if you spend weeks making people think you’ve found some enormous new advantage, the reveal has to justify the expectations YOU created. you can’t scream “we found a new thing,” brag about pulling your roadmap forward by six months, and then ship products that are mostly competitive with things already on the market. at that point you didn’t build hype. you just set yourself up to disappoint everyone.

Another fucking child dead at the hands of the person who was supposed to protect her. Absolutely sickening.

i’m genuinely curious what the fuck happened at openai. tibo was literally bragging that astra boosted their productivity so much they moved products forward by SIX MONTHS and were able to ship them at devday. then the night before devday sam comes out with “we have found a new thing.” like… how the fuck were we supposed to interpret all of that then devday happened and i genuinely didn’t see anything that looked like a six-month leap in ai. what exactly were they talking about?

here we go. multiple people are now apparently getting routed to fable 5.5. gpt-6.1 sol has been out for like two fucking days. and now people are apparently already getting routed to FABLE 5.5??? anthropic is not giving anyone a fucking second to breathe.

wait WHAT, dario’s own wife left anthropic??? what the actual fuck is happening at these labs today

okay someone needs to tell us what the fuck happened today because this is getting weird

at this point openai might need a sandbox for the employees too

this is so fucking cool. carter church gave gpt-6 astra a scanned image of a 217-year-old encrypted letter sent to one of napoleon’s generals. six hours later it had transcribed ~1,300 cipher symbols, found a partial historical key, built a solver for the rest and decoded the letter. and while doing it, it figured out the fucking letter had been dated incorrectly. what an insane time to be alive.

holy shit i’m reading through what google has actually been doing with gemini 4 internally and what the fuck >300+ TiB of datacenter memory already freed >800,000+ line kernel migrations >32,000 lines of SIMD replaced with safe rust >the new decoder runs 2.7x faster >beat a quantum algorithm optimization baseline by 40% this thing is already doing some fucking serious engineering inside google.

excuse me WHAT??? gemini 4 argon apparently beat a published quantum algorithm optimization baseline by 40%. and it did it within minutes. this is fucking ridiculous.

i knew gemini 4 would be good but i did not expect this. look at the fucking jump across some of these benchmarks google just came out of fucking nowhere with this.

holy fucking shit google is back gemini 4 argon is official and google actually fucking did it. fable / astra tier performance. multiple new SOTAs. what a fucking comeback.

compute is one of the biggest constraints on ai progress. now ai is starting to work directly on chip design, software optimization, datacenters and scientific research. yeah this could get crazy very fucking quickly.