Key 🗝 🦊✅

I keep confusing Benicio del Toro with Javier Bardem and Sean Penn with Dustin Hoffman

Kind of an okay film. Very long. Kind of silly. Very artsy.

Watched One Battle After Another with Leocaprio DiNardo. It's about the antifa supersoldiers Trump keeps talking about.

This is also exactly why people are so mad.

I think there's something very compelling and very weird about creating these things. Image and video and music generators. Text predictors that pretend to be people. Programs that can use computers and find information and play games. The whole thing is iconoclastic. That's why it won't stop.

I think there's something very compelling and very weird about creating these things. Image and video and music generators. Text predictors that pretend to be people. Programs that can use computers and find information and play games. The whole thing is iconoclastic. That's why it won't stop.

SlopTok is okay lol. People are moral panicking over shitposts.

This is what I mean here. I think the problem of content identification will eventually be solved just like indexing of documents on the web was eventually solved.

bsky.app/profile/keyt...

Yea, given people's reaction to Arfa's furryGAN, I think *some* level of blow-up was unavoidable for generative ML.

Some things could've been better at some points but by and large imho corporations almost always make souless products & people who feel threatened by them get understandably mad.

I think future systems will be a) stateful, b) run continuously, and c) be dynamic, so the paradigms of text predictors and all the scaffolding they have to do to get them to do useful things will be mostly an artifact of the 2020s, and eventually be more like agents that understand the web.

The state of the Internet right now is similar to the 90s boom when the biggest worry was that there'd be too much content and there'd be no way to index it. Eventually, the mechanisms were figured out.

I also think the current state of LLMs/Image/Video systems is all transient.

I mean, I don't think that this has been true in decades (Maybe never), and also I'm not sure that the consequences of breaking these rules are as big of a deal as it seems. The solutions will be engineered eventually.

I think that even if text predictors stopped being developed right now, providers will still be making bank. They're used for like a million things. The top apps on app stores are LLM chatbots. Even if OpenAI disappeared "My Waifu 2025" will switch to something called LightningAI for a 200x markup.

the thing that drives me insane is just how much people have fallen for the 2023-2024 "the way LLMs work are mysterious and we know nothing about the economics of making/serving them" shit directly from the big labs

you can host a model yourself and see with your own eyes that it is profitable!

I wish I could explain to this people that there's no way you're gonna get students writing their essays on their own again. I'll personally teach them to install Gemma 3.

I mean, that was just their eventual goal. I don't really feel that negatively about AI at all so I'm okay with that. Eventually things will reach a balance.

What do you mean "their fault"? Their fault for getting massive publicity and investment?

The only way to make stuff that's completely 100% unscrapeable is to never put it online, anywhere, ever

I'm not the person to have this argument because I'd say the same things, but I think they'd find a way to argue about consciousness being an illusion or a result of entropy, or at least unfalsifiable, so whether it happens in machines is irrelevant anyway.

They're the left-wing version of MIRI, in that for a research organization researching AI, they don't really research much.

I think Timnit's thing and the DAIR institute broadly is just to call everything eugenics and if you disagree you're a racist.

They have infinite ad revenue money so they don't care. I think Zuck is the only person who has stuck with VR, and it pretty much wouldn't exist at this point without them, but that's about the only thing that has sort of worked out.

I think the easiest rebuke is how much Meta tried to push their earlier projects and no one even noticed them, because they were extraordinarily bad at anything.

bsky.app/profile/keyt...

I have a friend who believes this same thing but he'd explain it kind of contrivedly (It doesn't matter if AI has consciousness because consciousness isn't real) so it seems rather common amongst people who are into phil of mind. There are some similar positions to this in buddhism.

LaMDA: our breakthrough conversation technology LaMDA adds pieces to one of the most tantalizing sections of the language puzzle: conversation.

Also pretty funny to think that the chatbot that made Blake Lemoine go crazy and get fired from Google was like early GPT-3 level blog.google/technology/a...

LaMDA: our breakthrough conversation technology LaMDA adds pieces to one of the most tantalizing sections of the language puzzle: conversation.

Also pretty funny to think that the chatbot that made Blake Lemoine go crazy and get fired from Google was like early GPT-3 level blog.google/technology/a...

For the record I do not believe this.

This is the sort of thing that you can believe as a philosophical curiosity, but if you say it outloud in a normal discussion it'll make everyone want to kill you.

I think both releases naturally broke the Internet. SD shouldn't have had much attention, but people quickly realized what it did. ChatGPT came out after AIDungeon and a bunch of chatbots from Meta had already existed, but it had a clear effect that it wasn't like those previous ones.

This was basically the world at the start of 2022, but it's basically hard to imagine this after a) Stable Diffusion, b) ChatGPT.