I can't imagine living in LA on $60k/year. $30k I'd just assume is living with parents, 4 other roommates in a tiny house, or they are just supplementing while their partner is the breadwinner. Oof.

Maybe. Generating even a single image of that size takes a long time right now, though, let alone enough images to fit a decent length video.

You could create dozens of unique characters per day, but ultimately they're going to be linear. They won't react in-the-moment based on your/other NPC's decisions. A future version of GPT could allow for real-time decision-making by the NPCs. Those items and such? All generated on the fly.

A lot of people are scared about the repercussions, but personally I'm incredibly excited. Yeah, there are going to be bad actors that use it maliciously. There's no way to stop that; even if it takes a large company to run a large model that can restrict biases and such right now, a few years from now anyone will be able to do it on their own PC in 2 seconds.

But the beneficial uses? Lordy. Even skipping over the basic stuff like photoreal pictures/videos, once we can combine all of them is going to be amazing. Imagine an RPG game where all of the dialogue, quests, etc. are crafted procedurally. "Completing" the game won't happen in a week at that point since you could have literally thousands of different quests with procedurally generated dialogue, storylines, and NPCs. You may not even be able to "complete" it at all.

On the extreme end, I see an entire world (see: MMORPG) being created by AI before being fleshed out/detailed by humans with a few hand-crafted elements for details the devs specifically want. Characters/art being generated by AI. Dialogue/writing elements being generated by language models. Entire libraries of AI-generated books based on the in-game world's history. Or even skip those last few things and just give a baseline world and let it run for 1,000 years in-game time to see what comes up, i.e. let the world detail itself. Bards at the local town tavern playing AI-generated music. The biggest thing stopping this from happening right now, I think, is finding a way to make all of these work together in a coherent way. Toss in an AI that knows how to code/optimize stuff and it's even better.

>Then comes the real world body avatars

RealDoll is already playing with that by having AI in porn dolls, e.g. they react verbally based on what you are doing.

I've seen various estimates that there will be newer language models 10x better than ChatGPT within the next year. The speed that the field is improving is absurd. I'd be unsurprised if we get low-res videos by the end of this year (by low-res I mean 480p maybe). There was already that 24/7 AI-generated Seinfeld on Twitch, although that wasn't exactly what I'd call quality.

Seems very high. Does it provide the median?