• What the article does not even mention is the increasing strength of the much cheaper Chinese OpenWeight models like Qwen.

    To be clear, if a genie in a bottle would grant me a free wish, I would request that this technology disappears from existence at least until we have solved the climate problem, found a solution how to compensate creative people and artists, and remedy its impacts on democratic discourse and mental health. But while one is allowed to wish for a better future, this is not the reality we face.

    I was requested at work to try out one of the local open weight models yesterday. They are not free, requiring substantial hardware costs, but require less energy than the US models. And it seems with a more modest and more sane approach, companies are not going to pay many hundreds of dollars per month per seat for Claude licences when they can do with far smaller costs.

    True that the “weaker” models will still require substantial and qualified human oversight for things like software development, but fully autonomous systems are more like a febrile hypercapitalist pipe dream than something technically desirable.

    • 2 minutes

      This is exactly it.

      When you run your AI in the cloud you are trading some privacy and security, as well as higher operating expenses, for convenience and lack of upfront cost. This is no different than any other cloud service.

      Go back a decade and everybody said cloud is the only future, companies will throw their servers in the garbage, etc. Now a lot of companies have moved workloads back on premises because the simple fact is the cloud provider is buying the same servers and has to pay the same price for them you are just paying for that and also for their profit and overhead.

      There’s also the simple fact that even the best models have proven unsafe to be turned loose without human intervention. So if you have to have a human reviewing the output anyway, why pay a boatload more for slightly better output?

      I just hope when this bubble pops there isn’t some taxpayer bailout. Let OpenAI go bankrupt if that’s what it takes. Let the idiots who financed data centers with blank checks lose their shirts. There needs to be a lesson for the next bubble.

    • 10 hours

      Fully agree. Having been in the tech business for a week or two thousand, I am always surprised that the fundamental truism isn’t widely distributed: cheap human assistance beats expensive self-driving every time.

    • Does it?

      Reuters is reporting that Anthropic’s Investor Day event a few days ago, where they showed their audited Q3 financials to investors, caused institutional investors to shift their sentiment and that they are being pushed to lower their IPO valuation target down from $2 trillion to $1.5 trillion.

      Coreweave is down almost 20% this week. Nebius is down more than 15%.

      Investor sentiment is shakier than it was a month ago. And for hype-fueled valuations, investor sentiment is a two-way street.

    • 5 hours

      2008 it was clear the market was going to implode, reporting came out by at least march of that year, around when I read about it. The mortgage market continued to boom, as the rich tried to offload their high end positions that were going to fall. (Their big scheme was to privatize social security, have the feds borrow 13 trillion or something, and force people to use investment advisors to put that money into the soon to fail market to allow them to cash out on a high note. Citizens would realize all losses but only keep up to 8% of any gains.) The Market didn’t fall until September of that year.

  • 15 hours

    Well if you couldn’t afford a home right now you could make a killing in buying up some of these debts, just make sure you get it insured.