SILICON VALLEY - The CEOs of OpenAI, Anthropic, and GrokAI have released a statement agreeing that claiming artificial intelligence "could potentially kill all humans" sounds much more awesome than just admitting that they blew a fortune on glorified chatbots.
  • Thorry@feddit.orgEnglish
    2·
    4 days ago

    OpenAI is crashing out hard. Their new models are marginally better and much more expensive. Customers have started to notice the models do pretty much the same thing, as the difference is only shown in some benchmarks (which may have had the Volkswagen treatment). So they question why it should be so expensive.

    The era of tokenmaxxing is over, companies are starting to look critical at costs and roi. OpenAI took those tokenmaxxing months, extrapolated that to a full year and told those numbers to investors to get more money. Now those same investors are starting to worry those yearly numbers might not materialize. Especially since Anthropic and Google appear to be doing better.

    Them falling out with Microsoft has left them much worse off. Sure they are more flexible now, but Microsoft was their biggest partner. Now they still need to pay Microsoft and Microsoft is free to run their own models for Copilot. These models are cheaper for them to run and more suited for the kinds of things Copilot is used for.

    They pay their execs huge amounts of money, actual obscene amounts. Yet one after another exec leaves the company. Some for valid reasons, some because they ran away after seeing the real numbers presumably. They consistently refuse to show any real numbers, instead doing a bit of creative accounting and marketing BS. When they go to market, they need to show the real numbers. This has proven to be such a hurdle, the IPO has been called off for now. They are claiming next year, but with the IPO of SpaceXAI and possibly Anthropic this year, the bubble might bust this year. Or at the very least the market will be weary after losing money on SpaceXAI and most likely Anthropic, given the rug pull nature of these IPOs.

    At the same time the spending is up, their commitments are mounting and coming due. The previous round of investment is drying up, so if they don’t do something soon they will run out of money within months. They can always go to good ol’ uncle Huang for a handout, but that’s not an infinite source and Nvidia would likely start making demands in return.

    So they’ve announced a pause because of ‘safety’, which makes no sense at all. They’ve thrown safety to the wind thus far and now suddenly stop. They claim they have some monstrous new model that is Skynet with an OpenAI badge slapped on. But they have claimed their new models are amazing for years now. Each time when people actually got to use the thing, the reaction was very much “meh” and they had to scramble to get an update out the door to appease the users. They would also need a huge breakthrough if they want to significantly improve the models. The things they’ve been doing till now have run into diminishing returns hard. There haven’t been any papers or even rumours this is the case, so I doubt it.

    More likely this pause is a desperate attempt to cut some of their spending. Cutting what they can where they can. As SpaceXAI has cheap and dirty datacenters up and running and no actual users, they have been dumping compute onto the market for cheap. Free from their leash from Microsoft, OpenAI will buy some of this compute in a further effort to reduce spending.

    We will see if they can get themselves into a more stable position, or if this is a death spiral. If they crash out someone will buy the pieces I’m sure. And it won’t be the people responsible left holding the bag.