Notices by Larry O'Brien (lobrien@hachyderm.io)

@flexghost @Techronic9876 Do bluechecks get a cut of ads shown near their tweets? What is “ad-revenue split” on #birdsite ?

@thomasfuchs @lisamelton Citation needed. The differences between GPT generations have been qualitative so far. More blocks and larger context windows mean more abstract features and more state. Don’t know how you conclude that won’t improve quality. The “tech bro” conceit is “scaling is _all_ you need,” following “The Bitter Lesson” argument: meta-methods that can find and capture complexity > methods inserted manually. Don’t know any studies that strongly refute this.

https://www.cs.utexas.edu/~eunsol/courses/data/bitter_lesson.pdf

@thomasfuchs @jkbecker @eniko My claim is that the limits of connectionist architectures are not evident, at least to one person who “knows the math.” I can’t prove that a connectionist champ-poker model is possible, but I’d bet on it. And I’d bet that, were such a thing to come out, some would confidently say “There’s no real concept of bluffing in the weights.” They seem to *define* “understanding,” “creativity,” “conception,” as involving qualia. OK, but what’s vocab for zombie-capability?

@jkbecker @thomasfuchs @eniko “Vast sol’n space” larger than the # of atoms in the Universe. Algorithms aren’t known for even professional play, much less championship. AG0 doesn’t even start with a “book” of openings and endgames. “Move 37” was outside professional experience. If we don’t believe these systems “discover” anything, if the trained models can’t be said to “understand” the rules and patterns of success, what’s the more accurate vocabulary?

@jkbecker @thomasfuchs @eniko True, Go can be played by a machine internally at vast speeds with perfect information and therefore you might think it’s dismissable as a “mathematical game” (Go enthusiasts would vehemently disagree). But poker, with only imperfect information available, is harder to similarly dismiss. Libratus — not a connectionist architecture, true! — does champ-level H2H Texas Hold ‘Em. Would you contend a connectionist Libratus-alike is impossible? On what basis?

@thomasfuchs @eniko I totally agree about nothing being close to *universally* near human. But if you can play championship Go with "no creativity, no deeper insights," then... are we sure we're correct about what those words imply?

@thomasfuchs @eniko They're connectionist architectures and don't compute in anything like the manner of, e.g., Deep Blue. They aren't performing search or brute force plies. They demonstrate "it's just statistics and remixing" may not be as scathing as it's often presented.

@eniko @thomasfuchs Would you argue the same about AlphaGo or AG0? In those models, the "hard ceiling" seems above the best humans can achieve.

How do you GM/DMs manage loot the #ttrpg players haven’t appraised? Just keep your notes until they do? Just tell them the value ‘cause life’s too short? QR codes? #DnD

I 100% believe this story

https://ludic.mataroa.blog/blog/i-accidentally-saved-half-a-million-dollars/

@thomasfuchs I'll bet you a nickel that site is styled with Tailwind.