Login
You're viewing the functional.cafe public feed.
  • Oct 24, 2026, 3:39 AM

    @icculus I will not predict any specific outcome with regard to AI capabilities. clearly they can produce words, images, math, sounds, videos, code, music, and various other kinds of Content, to some degree of efficacy.

    however I do have both a logical and an empirical coda to this uncertainty.

    💬 1🔄 0⭐ 0

Replies

  • Oct 24, 2026, 3:39 AM

    @icculus logically: These are plausibility machines; their outputs are by construction always more plausible than they are accurate

    empirically: every substantially impressive result achieved by such a tool has, after months of analysis, turned out to be either entirely fake or far more limited in scope than it initially appeared. but it does take months to figure out where the gaps are.

    💬 2🔄 1⭐ 0
  • Oct 24, 2026, 3:39 AM

    @icculus the nature of the plausibility engine is that every heuristic that you would use to initially determine whether a result is good, or economically valuable, or sufficiently carefully constructed, is now broken. the outward trappings of a finished system are now trivial to fake. determining what is broken is now a major project instead of a quick glance.

    💬 1🔄 3⭐ 0
  • Oct 24, 2026, 3:39 AM

    @icculus so don’t despair. it is still incredibly unclear whether or not it is actually any cheaper to burn the unsubsidized tokens to do something like this than it is to just hire somebody on fiverr to do it for you at a modest contract rate. even if it looks kinda like they already did it, they have to prove it now.

    💬 0🔄 0⭐ 0
  • 💬 0🔄 0⭐ 0