enframed wrote: Sat Jan 13, 2024 9:27 pm

Compared to those incomprehensible movie scripts AI were generating a year or two back this is insane progress. It surprises me that this isn't something ghostwritten by an actual comedian and then simply voiced by AI.

It's also possible that's exactly what it is. But it's getting hard to tell. Apparently a comedy AI called "Dudesy" is responsible but who knows how much human intervention there was. There's only one AI George Carlin special that's been released, not ten thousand.

But I guess that's not the point; every practical application of AI is likely to have a human component. Anyway based on the part I checked out it's coherent, the delivery is pretty good, and it's not funny at all.

AlphaFold 3 revealed.

So, the future is here. Forced AI imposed on EVERY FRICKING EMAIL I get in my work Outlook.

What does this mean, exactly? It means a gigantic "button" (actually a banner like a website ad) that eats up a ridiculous amount of space at the top of my reading pane which says "let CoPilot summarize the email for you". No way to disable that badboy...ohno...wouldn't want you missing out on that burst of increased productivity.

So I ask you, how smart is the AI solution deployed by the worlds premier software company exactly? So smart that it can't figure out that no one would ever want it to summarize an email that is one sentence long, Please, just nuke me now and lets get this over with.

A_Man_Who_Tries wrote: Wed May 08, 2024 10:43 am AlphaFold 3 revealed.

See you post things like this, which admittedly I don't watch for more than a few minutes because 1) the dude annoys me, 2) I don't understand some things, and 3) I don't care.

Then I listen to a podcast that says that these large language models get the answers to questions like "What time is it in London?" incorrect. What's a motherfucker supposed to think about all this?

enframed wrote: Tue May 14, 2024 7:31 pm See you post things like this, which admittedly I don't watch for more than a few minutes because 1) the dude annoys me, 2) I don't understand some things, and 3) I don't care.

Then I listen to a podcast that says that these large language models get the answers to questions like "What time is it in London?" incorrect. What's a motherfucker supposed to think about all this?

They're doing two different things and, as annoying as the papers guy is, the videos are always well put together and nicely explained, and worth sticking out.

Advances in protein folding, via these models, has already had a profound social impact when it came to sussing out the first Covid vaccines, and it's doing similar work with the rest of our major health conditions.

And let's have it right - buy a watch and do the maths.

enframed wrote: Tue May 14, 2024 7:31 pm Then I listen to a podcast that says that these large language models get the answers to questions like "What time is it in London?" incorrect. What's a motherfucker supposed to think about all this?

That you don't use a LLM to find out the time in London. That's like trying to do open-heart surgery with a socket wrench.

People have been saying they wished you could talk to a computer like it was regular folk for the last fifty years and it was always impossibly unrealistic but now you finally can, and people just go "ok, well that sucks". As somebody who was fairly well-acquainted with the prior limitations of computers, the fact that any of this stuff works AT ALL feels like mind-blowing sci-fi shit. And we're only just beginning to see what neural networks can do.

After a long while of being basically the same thing but a little bit faster, computers are getting freaky again.