Yesterday was a big day for me; my whole career as a web dev I’ve pined after the sorts of wildly intense and polished sites that win webby’s and so forth. And yesterday, my many MANY late nights and feverish slack statuses paid off with both this from FWA https://t.co/dZDTsz43Zp
I thought it was pretty clearly explained in the exposition between the pair in the dining room: the place "remembers" things, but each time it "remembers" them, "it does less." So in some sense the Pirate Clark copy/memory is (not to be too punny about it) more fleshed out, being that it's a fresher amalgam of him. Whereas as other copies are no doubt copied from copies of copies, they lose more and more fidelity/depth until what should be skin and muscle and blood and bone becomes simplified/ignored till it's just rubbery almost-skin over moist white mush (though still edible); or the correctly spelled signs become incomprehensible but in related font (or even mirrored like the Stop sign from the beginning).
So for instance, that's why in the end they were showing all the MRIs: those MRIs were of Mary for the scientists to test whether she's actually "complete" or original and not just a shallow copy somewhere under the surface.
Considering all the phrases highlighted by Pangram as particularly indicative of AI generation _were in quotations from the very AI generations quoted for the sake of the piece’s criticisms_ I’m actually not terribly confident Pangram’s system is sufficiently sophisticated to avoid a possible false-positive like this.
Y’all. 👏👏👏
(Fyi, this is also exactly my experience of being a frontend dev — since the mid-2000s at least — and a coder more generally — basically my whole education and career.)
(Relatedly, I’m convinced there’s a Moore’s Law of technique when it comes to computer science — where the effort required to create a thing of a given complexity exponentially decreases over time and/or accomplishes tasks of exponentially more complexity — and that that phenomenon is ultimately just a higher order or branched expression of whatever deeper, mysterious mechanic it is that results in the textbook Moore’s Law in the relationship between computation power and enduser cost.)
@EvanAKilgore There’s nothing cheap in pointing out Erika’s hypocrisy decrying others’ “hundreds and thousands” made per episode, while dabbing her way to $140,000,000 in just 3 months.
And some things are mutually exclusive: grieve OR lead. When you’re doing both, you’re doing both poorly.
@WR4NYGov@KonstantinKisin You just missed the joke. It’s that the checklist to make it big in the new media is empty. (Totally breathtakingly brilliant way to end the article.)
@AbeChomali@KonstantinKisin (You just missed the joke; that last part wasn’t behind a paywall, it’s that there’s nothing at all concrete or substantive required to be a prominent member new media.)
(Truly, truly brilliant ending)
@levelsio recall. Hence “memories changing” as your LLM trains very different data (changing memories) or much more data (remembering additional, possibly false, possibly true, details as you mature). 5/5 + 2/2 + 3/3
@levelsio your image/experience conjuring functionality remains reasonably intact (aka consistent with the system when the low-discrepancy description was calculated), the lowest-discrepancy description of images/experiences should reliably allow for reasonably useful 5/5 + 2/2 + 2/3