GPT-4.5 is ready!
good news: it is the first model that feels like talking to a thoughtful person to me. i have had several moments where i've sat back in my chair and been astonished at getting actually good advice from an AI.
bad news: it is a giant, expensive model. we really wanted to launch it to plus and pro at the same time, but we've been growing a lot and are out of GPUs. we will add tens of thousands of GPUs next week and roll it out to the plus tier then. (hundreds of thousands coming soon, and i'm pretty sure y'all will use every one we can rack up.)
this isn't how we want to operate, but it's hard to perfectly predict growth surges that lead to GPU shortages.
a heads up: this isn’t a reasoning model and won’t crush benchmarks. it’s a different kind of intelligence and there’s a magic to it i haven’t felt before. really excited for people to try it!
@bindureddy Proof? Human General Intelligence never makes things up when they don't know the answer.
Have you observed children as they develop? They make the most interesting statements about what they believe.
Check out the split brain experiments from Caltech...
@ID_AA_Carmack@jasonbotterill When I talked to the CTO about paying for the subscription. He asked how much my time was worth. I realized the one question I asked had saved me about 2 hours and already paid for itself. I subscribed moments later.
I made this reply to one of the first “ChatGPT would find that bug in seconds!” comments, but 300x more people saw the original post, so there were a dozen more similar comments.
A little more context: the code was using PyTorch, where ge() is greater than or equal, and gt() is greater than. The code executed fine, and the loss was even decreasing, just doing much worse than it should have. Being all tensor ops makes inspection harder, and I was also fighting obscure multi-GPU internal issues at the same time. Of course, in hindsight, I can imagine much better strategies that would have found it quicker, but I don’t feel foolish about it.
@nicolefv Agreed @nicolefv, but I've come to realize that most people just learned their boundary behavior growing up. Unfortunately that leads to people who don't know what healthy boundries are, even if they think they do. Time for an international standard?
@samkrishna@nicolefv I'm going with people who have the audacity to take incomplete information and instead of asking for more information, decide they know and force their view on everyone... like this... she was clearly talking about loving the audacity audio program! @nicolefv, I give up, context?
@laceynwilliams I was working remotely for 6 years, got good at it, and decided to get back in the office... got a new job, then Covid sent me back to remote. What I learned: 1) Remote gets easier over time and 2) After 6 years I missed people.
@elonmusk 1) I am sure you're right, but 2) without the context in your head, I expect people won't behave differently.
Why is it dumb? What should people do differently and why? @elonmusk
@royalsociety He was awesome. I squared R heating is why room temp superconductors aren’t needed to carry power long distances efficiently. Just jack up voltage.
For #GivingTuesday, help us put food on more tables over the holiday season.
For every RT of this post, we’ll make a donation to help @FeedingAmerica provide 10 meals* to families in need. #GivingOnUs
📲: https://t.co/UwVsfEJewL
@NandoDF In an ideal world, all research would be published, even with negative results. Maybe the solution to "too much reviewing" isn't to tell people to publish less? Instead maybe we should seek ideas for improving that process?