Who Is Anthropic Kidding?

(theatlantic.com)

20 points | by nothrowaways 16 hours ago ago

9 comments

  • simgt 15 hours ago

    I feel like we could stop training large frontier models now and still have a transformation of at least the scale of the internet on the economy. We still have extremely bright humans to push the boundaries of science, faster than ever before now that any of them can do simulations and data analysis for a few tokens. Anyone arguing for more is playing against us, either by being naively optimistic or because they're convinced their wealth will shield them from the consequences.

    • nixosbestos 13 hours ago

      What is even the point of this line of thought? As long as GPUs exist, someone out there is going to keep trying to advance and train new models.

      There is no putting the cat back in the bag.

      • simgt 8 hours ago

        We have heavy regulations and international collaboration on genetic engineering or nuclear. I don't see how that's any different. Training SOTA models now requires a lot of capital and hardware which makes it harder to conceal.

      • dieselgate 13 hours ago

        I think Parent argument is more along the lines of the 80/20 rule where the major advances are already commercially available.

  • enjoyyourlife 8 hours ago

    We're so spooky! Buy our stock! I think they just want those government contracts.

    Previous discussion: https://news.ycombinator.com/item?id=50518323 (Just my comment.)

  • TimSchumann 12 hours ago
  • cyanydeez 14 hours ago

    Themselves.

  • A_D_E_P_T 16 hours ago

    "Has to be me. Someone else might get it wrong."

    That's their attitude, and I don't think they're hiding it.

    • cajal 12 hours ago

      stop imagining these clowns as having anything resembling morals. it’s not even that. it’s:

      “Has to be me. Someone else might make that money”