Comment on ChatGPT to start showing ads in the US

• definitemaybe@lemmy.ca ⁨1⁩ ⁨day⁩ ago

  OpenAI has a snowball’s chance in hell of surviving. LLMs cost a lot of compute. They’re burning through cash. Operating costs are more than double revenue. Their net operational losses are about $1 million USD every 40 minutes.

  And somehow they’re trying to put half a trillion USD into building more datacentres to make even more advanced models, which will be even more compute intensive.

  Meanwhile, as venture capital has been committed to a whole series of AI companies and data centres, venture capital is dying up. Nobody has gotten a payout yet, since there’s no path to profitability for any of these companies.

  And it’s coming to a head this fall, when OpenAI needs to pay their suppliers for the expansion they’re building, and there’s no reason to believe they’ll be able to raise enough more investment to cover their costs.

  It doesn’t even take OpenAI failing, either. There’s so much debt (“leverage”) and circular cashflow going on in this space, between the AI companies, data centres, computer hardware manufacturers, and construction companies, that any one of them failing could cause cascading failures, like dominoes. Worse than the '08 financial crash, most likely.

  So no. It’s not going to be like YouTube. YouTube is cheap to run, compared to LLMs.

  And the worst part of it all: LLMs aren’t even very good! It creates an illusion of productivity, but it’s all bullshit, either doing a shitty job, or taking more time to prompt fondle than it would have taken to do the job by hand, or building up tech debt that’s going to make massive projects unmaintainable.

  It has some use cases, sure. I use it almost daily, tbh. But only because someone else is footing the bill. It doesn’t produce nearly enough value to justify its costs.

  source

  • Powderhorn@beehaw.org ⁨19⁩ ⁨hours⁩ ago

    Don’t kinkshame … some of us enjoy prompt fondling.

    source