flywheel in action 🔁

what he said!

if agi renders me obsolete you’ll know where to find me…

love seeing companies build on top of OAI models to tackle hard, domain-specific problems. I had an awesome convo with Omer Goldberg about how Chaos Labs is layering their own approach on foundational models to drive real, tangible value across the blockchain!

🚫 on imagegen refusals we hear you on the current state of refusals not being up to far (amidst the transition discussed below)! favor: can you please reply with screenshots of imagegen requests that shouldn't have been refused, but did? that'd help us accelerate fixes. 🙏🏼

the real interface of AI isn’t a chat UI –it’s a feedback loop 🔄 metrics are easy. actionable insights are hard. flywheels that compound are built on: • high quality signals • fast, fresh & cheap exports (infra that scales) • 'frolicking in the weeds' • evals, evals, evals!

SF to Sausalito on a saturday morning. light fog. two wheels. quiet streets. phone on DND. my idea of luxury.

when I’m deep in a problem I care about, I get tunnel vision —but the vibey, good kind.

almost always, a company goes from model to moat by way of evals. if you can't measure your competitive edge, you don't have one.

ICLR in Singapore this week has been amazing —tons of interesting posters, talks, and workshops. Just as meaningful has been catching up with old friends and ex-colleagues. Great seeing Vahab Mirrokni , my longtime research partner —our healthy competition (and friendship) pushes

we talk a lot about big launches, but I love the quiet, relentless updates that improve 4o

We’ve rolled back last week's GPT-4o update in ChatGPT because it was overly flattering and agreeable. You now have access to an earlier version with more balanced behavior. More on what happened, why it matters, and how we’re addressing sycophancy: openai.com/index/sycophan…

our model launch process is continuously improving and we learned a lot from this!

our kids will think we were crazy for using Google for 20 years "so you typed a question, got 100000 blue links gamed by SEO agencies, opened 11 tabs, skimmed each site, pieced together an answer, and repeated this process 20 times a day?" they'll think we were digital cavemen.

AI jammed into an existing interface < an interface designed to maximize user's leverage in a specific domain. nice read Pete Koomen koomen.dev/essays/horsele…

who grades the graders?