Yacine argues the unexplained capabilities of deep learning are more concerning than theoretical AGI risks

@yacineMTB what does it even mean to be AGI pilled but not neural network pilled

I remember watching deeplizard videos 8 years ago, training a Mario sprite to play Mario, and thinking "the future is now old man!"

Just looked them up rn and after being absent these past couple years they just uploaded a video 2 weeks ago where they're in a forest stating they've moved to isolate themselves up in the Appalachian Mountains and distanced themselves from AI, though they still believe deep learning to be highly relevant.

@HumanClanker @yacineMTB @eigenrobot see now we -could- have gone the Dr. Chandra route and built HAL from the organic gel pack shit

(which i mean 20yrs from now who knows)

using clunky regular computers seems to have been safer

@Clipart_Bear @yacineMTB At a time when the Brits weren't particularly fond of us, but they really liked the guy

@yacineMTB this is like the old testament

@yacineMTB I’m so glad we live in a world where we got compute and text based models instead of getting a bunch of compute and then accidentally RLing an unknowable machine god into existence

@yacineMTB Can a neural network shit post on the internet

@yacineMTB Even without deep nets, a linear layer (literally a single matrix multiplication) solves a stupid amount of problems

@yacineMTB AGI is like a warm fuzzy hyper IQ bunny compared to 12 dimensional neural network mesoamerican demons

@yacineMTB how about you back propegate onto those job applications mr has a family to support. how about you navigate into the vector space where you have a cottage how about that

@yacineMTB Kinda narrow applications where they are really good, that's why there are boom and bust(ai winter) cycles People get v high off the "computer do everything" gas, and apply machine learning poorly, and ppl abandon it

@kitledru @yacineMTB Starting video you’d recommend? I went down this rabbit hole for a couple weeks and found it unbelievably fascinating, but don’t think I’ve consumed any content from the majority of these folks

@RasputinKaiser @yacineMTB Yeah a lot of people still hand write addresses, and the OCR can read and sort most of them automatically, but you still need a ware house of people handling edge cases manually and training the system to detect them better, but this goes on forever you can imagine, always edges

@Clipart_Bear @yacineMTB The postal system overall is just so...intense. I had read into it a bit after seeing about an American abolitionist who was really into the idea of making mail cheap and accessible. https://en.wikipedia.org/wiki/Elihu_Burritt#:~:text=Burritt%20advocated%20that,in%20his%20lifetime.

@RasputinKaiser @yacineMTB Yeah and it's kind of interesting we kept improving mail infrastructure alongside the invention of the telegraph. The pigeon and the horse have been retired but paper mail is still pretty load bearing part of our communications. Makes tech "replacement" narratives feel silly

@yacineMTB recommendation models are alien brains.

What human can sort a bazillion content posts a second and maintain personalization? Look no further than that to see what weirdness we can summon when pointing neural nets at niche problems and data

@Clipart_Bear @yacineMTB Luckily countries with different alphabetic systems are now making pretty sturdy OCR models, so I'm sure having more resources other than Latin is handy for edge cases!

@Clipart_Bear @yacineMTB The shit government agencies used to accomplish was leagues ahead of corporations. Now we sort of shudder at government tech and it's a shame

@Clipart_Bear @yacineMTB I always wondered how they processed so much mail! That's really cool, old government funded programs were insane before they were systematically enshittified and promptly defunded because they were shit from... interference and underfunding lmao