Need to let loose a primal scream without collecting footnotes first? Have a sneer percolating in your system but not enough time/energy to make a whole post about it? Go forth and be mid: Welcome to the Stubsack, your first port of call for learning fresh Awful you’ll near-instantly regret.
Any awful.systems sub may be subsneered in this subthread, techtakes or no.
If your sneer seems higher quality than you thought, feel free to cut’n’paste it into its own post — there’s no quota for posting and the bar really isn’t that high.
The post Xitter web has spawned soo many “esoteric” right wing freaks, but there’s no appropriate sneer-space for them. I’m talking redscare-ish, reality challenged “culture critics” who write about everything but understand nothing. I’m talking about reply-guys who make the same 6 tweets about the same 3 subjects. They’re inescapable at this point, yet I don’t see them mocked (as much as they should be)
Like, there was one dude a while back who insisted that women couldn’t be surgeons because they didn’t believe in the moon or in stars? I think each and every one of these guys is uniquely fucked up and if I can’t escape them, I would love to sneer at them.
(Semi-obligatory thanks to @dgerard for starting this)
oh hey, we’re back to “deepmind models dreamed up some totally novel structures!”, but proteins this time! news!
do we want to start a betting pool for how long it’ll take 'em to walk this back too?
it’s weird how they’re pumping this specific bullshit out now that a common talking point is “well you can’t say you hate AI, because the non-generative bits do actually useful things like protein folding”, as if any of us were the ones who chose to market this shit as AI, and also as if previous AI booms weren’t absolutely fucking turgid with grifts too
I suspect it’s a bit of a tell that upcoming hype cycles will be focused on biotech. Not that any of these people writing checks have any more of a clue about biotech than they do about computers.
Haven’t read the whole thing but I do chuckle at this part from the synopsis of the white paper:
[…] Our results suggest that AlphaProteo can generate binders “ready-to-use” for many research applications using only one round of medium-throughput screening and no further optimization.
And a corresponding anti-sneer from Yud (xcancel.com):
@ESYudkowsky: DeepMind just published AlphaProteo for de novo design of binding proteins. As a reminder, I called this in 2004. And fools said, and still said quite recently, that DM’s reported oneshot designs would be impossible even to a superintelligence without many testing iterations.
Now medium-throughput is not a commonly defined term, but it’s what DeepMind seems to call 96-well testing, which wikipedia just calls the smallest size of high-throughput screening—but I guess that sounds less impressive in a synopsis.
Which as I understand it basically boils down to “Hundreds of tests! But Once!”.
Does 100 count as one or many iterations?
Also was all of this not guided by the researchers and not from-first-principles-analyzing-only-3-frames-of-the-video-of-a-falling-apple-and-deducing-the-whole-of-physics path so espoused by Yud?
Also does the paper not claim success for 7 proteins and failure for 1, making it maybe a tad early for claiming I-told-you-so?
Also real-life-complexity-of-myriads-and-myriads-of-protein-and-unforeseen-interactions?
As a reminder, I called this in 2004.
that sound you hear is me pressing X to doubt
Yud in the replies:
The essence of valid futurism is to only make easy calls, not hard ones. It ends up sounding prescient because most can’t make the easy calls either.
“I am so Alpha that the rest of you do not even qualify as Epsilon-Minus Semi-Morons”
i suspect - i don’t know, but suspect - that it’s really leveraging all known protein structures ingested by google and it’s cribbing bits from what is known, like alphafold does to a degree. i’m not sure how similar are these proteins to something else, or if known interacting proteins have been sequences and/or have had their xrds taken, or if there are many antibodies with known sequences that alphaproteo can crib from, but some of these target proteins have these. actual biologist would have to weigh in. i understand that they make up to 96 candidate proteins, then they test it, but most of the time less and sometimes down to a few, which suggests there are some constraints. (yes this counts as one iteration, they’re just taking low tens to 96 shots at it.) is google running out of compute? also, they’re using real life xrd structures of target proteins, which means that 1. they’re not using alphafold to get these initial target structures, and 2. this is a mildly serious limitation for any new target. and yeah if you’re wondering there are antibodies against that one failed target, and more than one, and not only just as research tools but as approved pharmaceuticals
but but proteins! surely they’ve got it right this time! /s
(I wondered what you’d say when I saw this. I can only imagine how exhausting)