Someone observed that though the tools offered by the current AI company front-runners have some value, it's hard to see that value justifying their valuations - maybe, as with the internet boom, the first round will essentially fail, and the second wave will actually work/stick.
Yep.
NB: A transformer, in this context, is a component of a deep neural network whose whole deal is, basically, being sensitive to context and amenable to being usefully stacked up, which makes them good for modeling complex hierarchically structured information (like language and images.)
(Ideally, rolling one's own transformers.)
Also, I know it's low-hanging fruit, and perhaps best ignored, but... transformers? Why haven't I heard about, at minimum, Autoencoderbots?
It's never not the case that what was once cutting edge research becomes a useful exercise for advanced undergraduates. When I was in undergrad, this would be writing a chess program. Now I reckon it's creating a Turing-test passing LLM from scratch
Or like the husk of an insect whose insides have been liquefied and drunk by a spider.
The problem with being a motivated but particular reader is that you find one of the few books you're going to like and then you read it and then it's read. Re-reading, yes - Borges was mostly a re-reader, after all - but after a while a much re-read book becomes like an orange squeezed flat
Read this, it's good (and I'm very, very rarely pleased):
www.amazon.com/Down-Time-No...
I hadn't looked at this in a long time but it still amuses me and definitely captures a moment (i.e., pagers existed, and it was necessary to specify that a laptop had wireless.)
amazork.com
Oops, I mean "alleged tortious act"
An attorney for Big Oil, defending it from the kami, who are in fact AIs pretending to be nature spirits, sort of, and have finally understood the uselessness of data and the importance of narrative in making a case to the public.
"Some computer programs pretending to be nature spirits have accused my client of being the sole and immediate cause of climate collapse, with their tortious act being, as far as I can tell, upsetting the ghost of a dinosaur."
Who'd have thought Michael Bay would make such a contribution to Artificial Intelligence?
"Chest deep in darkest Cheever country," said Wintour. "Cos-playing as a disappointed WASP whom life has let down."
- One of my favorite lines from Forthcoming Book. I love all my sentences equally, of course, but some are more equal than others.
Vision: best efforts, but I have a reasonably good idea of the realities of development. Perhaps better to be zen/accepting from go, lest I take to drink.
Moltbook: Ai ya! Looks like a joke/publicity stunt, but a startling effective one. That said, one day... Hmmn. Actually, there are bits of Memory Atlas* that are essentially about that.
* - Final Title To Come.
Me, now: "AI now actually does more than nothing! It's both incredibly impressive and millenially empty. Transformers are important, but not for the reasons people think. In future, there will be memes/jokes of the form 'Oh noes! LLMS are here! Quick, let's fire employees!'"
Me in 1993 on my third day as a first year PhD student: "Wow, AI academia is a joke! This is all totally worthless and will be justly forgotten."
Me now: "Yep."
It is! Though an animated anthology series (a la Love Death and Robots) seems most likely. To try to hold on and insist on my vision for the screen version seems like a ticket to madness and failure - I (would) see myself more like a grandparent, supportive but not trying to make decisions
There's a point where one can't see the book anymore, but just, as it were, the code - how it functions as an artifact - with whatever impact muted or indiscernible. Then eventually one forgets and the book stirs to life again.
(* - New title to come.)
Development heaven - easy - the studio head calls you out of the blue to say the deal is greenlit, and how else can he be of service?
Development bardos and purgatories are trickier...
The process of trying to get Memory Atlas* onto screens is beginning. It will be fun to see how it goes, though of course the most likely outcome is development hell culminating in nothing, which begs the question... what other kinds of development afterlife/extra-planar states could there be?
I will! It currently looks like sometime between late 2026 and the End of Days. Well, okay, actually between late 2026 and early 2027, but there's an awful lot of lead-time.
Finished the substantive editing on the new book. There's more to do but mostly just copy-editing, all fairly unintensive, which means I'm now essentially a passenger, which is always an interesting state. I feel like the book's ghost is leaving my body - publishing as exorcism - Vade retro librum!
A book I'm writing offers in passing a solution to the Fermi Paradox, namely that biological life is self-limiting; within a few millenia of the invention of computers it goes extinct, maybe survived by AIs - thus this silence.
It came to mind randomly but I now suspect it's correct.
Oh well.
A tech-interview question I made up: "You're given an infinitely fast computer. Not merely very, very fast, but *infinitely* fast. What are the coolest things you can do?"
The (disappointing) answers are always 'graphics and cryptography.' Better would be, say, "experimental cosmogenesis."
A prominent entrepreneur in his fifties recently compared himself to Alexander the Great.
Alexander died when he was thirty three. If he'd lived would he have ended up crapulous, pitiful, even grotesque? Would his concubines have said, "Can this be the man who once was Alexander?"
thebookofsand.net
A song cycle inspired by the Borges story "The Book of Sand."
