Sam Altman wants your RAM — there are Compute Campuses to build
Sam Altman of OpenAI loves going on podcasts. These are always tightly scripted promotional exercises. On last month’s interviews with Politico and Vanity Fair, he went through the same list of talking points in both podcasts. But every time Altman goes on a podcast, he lets the mask slip and says…
Annons
Annons
Sam Altman of OpenAI loves going on podcasts. These are always tightly scripted promotional exercises. On last month’s interviews with Politico and Vanity Fair, he went through the same list of talking points in both podcasts. But every time Altman goes on a podcast, he lets the mask slip and says something horrible that appals normal people. This time it’s from Sam’s podcast with Stripe founder John Collison on Thursday 29 October. . Sam wants his data centres to have priority for computer memory, and not you plebs: [YouTube; transcript] A lot of people want local inference. And somehow, they assume that the problem is solved now. Because, we are going to have open-source models. We’re going to be able to run them on our computers. And so we’re not going to need all the stuff in the data center. But from a global allocation perspective, actually, it’s quite terrible to have everybody have a bunch of the world’s limited supply of DRAM sitting in their own computer. And, being used 5% of the time. You want maximum utilization in the cloud if you’re trying to deal with this very supply constrained environment Altman performed outrage that his comment was taken out of context — he was specifically talking about people running AI rigs at home. You should just outsource your work to OpenAI, see. It’s much more efficient. [Twitter, archive] But people were upset because Altman ignored the real issue — it’s a “supply constrained environment” because of you, Sam! Memory prices have gone through the roof, making all electronics much more expensive, specifically because the AI companies, and mainly OpenAI, have locked in all the RAM supply for the next few years. Even in context, it’s a ridiculous statement. Local models are negligible. They don’t have enough RAM to make a difference and it’s the wrong kind of RAM for Nvidia data center rigs. And the guys running local models are OpenAI’s biggest boosters and biggest excuse. Altman did rename his Twitter to “RAM Altman” for a while yesterday. [Twitter, archive; Twitter, archive] Altman’s quote about RAM got all the publicity, but the rest of the podcast is pretty amazing too. What if, right, self-replicating AI data centers? There’s very little, relatively little talk about self-replicating data centers. This idea that you can use AI to control robots, and run fabs, and build turbines or solar panels, and build data centers. But, I think we are… I think we have that within grasp. Also, AI can do the chip designs! Pretty good for a super auto-complete. What else could the AI do? But if AI is going to discover really deep novel bugs in hardware that require physics we didn’t understand, but all of a sudden means some assumption doesn’t hold. Then I think, people aren’t sure what happens. What? This is all nonsense. Altman is saying stupid things about areas he doesn’t understand. It’s as disconnected as the gibberish Elon Musk comes out with. The facts and details don’t matter — they’re just spouting sci-fi to sell you on the chatbot. Everyone really hates data centers and AI right now and wants them gone. A lot of Altman’s planned data centers will likely never be built — especially after data centers were such a key issue in Tuesday’s US election. [NYT] What’s the problem? Must be the name. So Altman wants to rebrand data centers as “compute campuses.” OpenAI told Business Insider: [Business Insider] Compute campuses better describes the scale and role of the infrastructure we’re building for AI. The great news is that Sam can say very dumb things on all the podcasts but OpenAI is still running out of money. But Greg Brockman insists OpenAI can IPO next year after all. Maybe if the chatbot alters physics for them. [Reuters] Video — Podcast