I assume it's big tech that has this weird ai they try to sell while the scientists are using different ai for real useful stuff, like the protein something I heard. Or at least that's what I'd like to believe.

I don't think image generators are really in the same category though. They'll have their applications but they're not going to be a fundamental change to society the way AGI will be.

It’s part of AGI and will be a massive shift. They are to video what punk was to music.

Google just released a video generator that is a ball hair away from perfection. The hallucination rate from their latest models is <1% and dropping you just see cherry picked screenshots.

Agi and image diffusion has literally nothing in common though?

imagine if a company said they're going all in on quality instead

The physics and perspective are still horrible, every video makes me want to vomit my brain

They certainly didn't help it by weighing it down with more and more ads.

Only sometimes, with enough generations you can already make indistinguishable videos for the most part. You’re seeing these mistakes because it’s amateurs spending $100 not professionals spending $10k.

Given current rate of pace Veo4 should be out in a couple months.

Well one of the big problems with it is it's never properly configured. One of the most annoying things that it does is that it generates tasks only when previous tasks are closed, in theory that makes sense but really the result is that you close a task, and then you have to go looking in the ticket queue for the new task it's just generated, so you can close that one too. Total waste of time.

Yes it does. It’s one component of a broader system. The ability to generate helps it interpret. An AGI might use a diffusion model to imagine scenarios, generate visual plans, or process sensory input.

Can confirm. The more you deal with people who have climbed to the tops of corporate ladders, the more it becomes clear that it's all vibes. It's all people telling stories to other people who tell stories about those stories.

The peter principle is wrong - in an oversized corporate structure, there is no upper bound for incompetence. You can keep rising for no reason, because after a certain point other people just trust that you know what you're talking about, and the people that know better work around you instead.

The people beneath you can't trust the people above you enough to explain the situation, the people above you don't really listen to the people beneath you anyway, and so plenty of middle managers just muddle through and constantly make shit up to justify their own existence, while everyone above and below is left in the dark about what's really going on.

Decisions are constantly made by people without any real connection to the consequences, and it shows. With the everything.

But higher quality ≠ more profits
AI apparently makes investors wanna dump in all their money tho

The two thing I use most, by far, from Google, is gmail and basic search.

Gmail, I'm looking to move away from it now, but I currently have every little addition to it disabled. Basic inbox and tags, no automatic filtering, no categories, no nothing.

Search, my browser is set to open the "web" tab with the query, no transformation, no summary, no "for you", no AI garbage, no "we thought you wanted video so there's only video in the replies". It still works fine.

Basically, none of what they added for years… maybe decade at this point, had held a glimmer of interest from me. It feels like this trend will continue. I just want something very basic that works.

It was actually a really good product, way better than Facebook, unfortunately if you have a social media platform that's invite only then it's never going to succeed. I really have no idea why they did it like that.

The AGI, by definition, will make something vastly better than diffusion model. That's one of the cornerstones of AGI, it will explode it's capabilities.

The statement assumes AGI must surpass all narrow models in every domain, but this is not a requirement by definition. AGI is defined by its generality across tasks, not by superiority in each specialized field.

I stand corrected.

I was referring to the demos shown at I/O, my brain seems to see them all as “flat”, it’s like a reverse magic eye thing. It reminds me of the uncomfortableness of a fever dream. I may be the exception tbf!