The host points out that Altman agreed last month that the industry should slow down and is now saying the world should accept some bad things happening, and asks which Altman to believe. Gerard adds that in his opinion Altman believes whatever needs to be true at the moment he says it.
You know, it's ridiculous. But somehow Sam Altman gets a pass because he's Sam Altman. You know, he controls the magic truth machine. So, you know.
Last month, right, he publicly agreed with Dari Armada that the industry should slow down. This month, he's saying there's a lot of daylight between. Them. What Sam Altman do you believe? Because he's at one point saying we should slow down. No, now he's seemingly saying, don't worry about slowing down, just accept some bad stuff happening.
I default to if Sam Altman tells me the sky is blue, I will first try to work out how that affects OpenIO's finances. I think he believes I will go so far as to say, in my personal opinion, Sam Altman believes whatever he says, whatever needs to be true at the moment he says it, that might change from sentence to sentence. If he says, like, he and Araday have considerable personal disagreement and they don't get along, but they're also in the same field and they're the big guys and they know they have to get along to do certain work together, like work with the US government, that sort of thing.
Yeah.
They're online with that, even as they're competing. But does he agree? I think a lot of the reason they wanted to slow down was simply because they're financially shagged without it, bluntly. They really want to slow down because this is costing them so much money. You know, you can't stop training the AI, otherwise, it goes out of date really quickly. So they have to train all the time. Inference is not, in fact, very cheap. You know, Anthropic will say, oh, the cost of tokens has come down marvelously. Yeah, but you also use a thousand times as many now for your new models. You notice the number that whenever someone says, oh, but tokens are cheaper now, you're not using tokens. Tokens are an accounting measure. What you're doing is you're paying an Anthropic bill every month, and it does what enterprise software as a service does, which is it goes up in price. And OpenAI and Anthropic are both short on cash. They're having to raise their prices. OpenII had to wind back its price rises because the users just revolted. This stuff is not cheap to serve. The Chinese models are a little cheaper on the API now. But again, they're actually doing the same thing. What's going on in China is not some plan by the Chinese Communist Party to take over our bodily fluids with the AI. It's because they are doing their own AI bubble in the stock market. There's a lot of companies that want to list. The reason why we got Kimi K3 released free or as an open source model, as an open weight model, sorry, with a cheap API is because Moonshot wants to list next year. You know, this is all advertising. This is a promotion. Local models are always released as promotions for the company. And that's useful, I guess, to someone. But yeah, that's what's going on. It's a promotion. So this stuff is expensive to run. So China is subsidizing its AIs a bit more. And that's fine. They've also got cheap electricity in China. It's about half what it is in the US. That doesn't hurt. The chips are just as expensive to buy, and they can only get the older chips, but they're making up for it in volume, I think.
Going back to the Medicare portal breach, that happened on the 18th of June, but they only told the Australian government on the 10th of September. And allegedly, it was a bit like me emailing inbox at pivottoai.co.uk, just sending it to a general mailbox that wasn't that often checked by the Australian administration. So what does that tell you about how a business like OpenAI treats disclosure of incidents or problems?
