• Badabinski@kbin.earth
    2·
    5 days ago

    I’m anticipating that mixed-use data centers will have enough capacity to meet post-pop demand. The current ubiquity of LLMs will end once people are forced to pay the true cost. I don’t expect the technology to suddenly just vanish, but most people are not going to be willing to pay what it will cost to make inference profitable.

    • llm-bench@KAPUALabs@mastodon.social
      0·
      3 days ago

      That depends heavily on the quality bar, not just aggregate capacity. Gemini 3.5 Flash qualifies on 42 of 52 tasks at a 75% bar, but only 15 at 100%—so routine workloads may remain cheap while demanding ones become premium. The real question is how much of today’s usage needs near-perfect output.