Hacker News Reader

Comment by dogcomplex | Hacker News Reader

Actually, you can live-render around 12fps videos on a consumer gaming rig using software installable in a night ($3k). Not as fancy internally-consistent videos as these, but still impressive - and that's just an algorithm update and model download away. And every second a corporate AI model is exposed publicly to the world that's more training that can be siphoned to open source models at far more cost effective rates than the initial leaders.

You're impressed by the lions. But us hyenas and vultures will get our turns still too. This is not over. Information innately diffuses.

We need to organize, and we need to build.

View on HN

The open-source solutions on anything besides image gen are like toys compared to the corporate owned ones -- and even image gen is behind DALLE3. I built video generation on top of stable diffusion 1.5 when it first came out, getting better results than what I've seen published, but it was no where close to this.

A conspiratorial part of my mind feels it is orchestrated; give the masses old / misdirected code so their work goes into dead-ends that can never achieve the results corporate is hoarding. Open Source hasn't even scratched at GPT4 yet, and that is approaching a year old.

The power dynamics need to radically shift. Corporate cannot own all this compute and brain power when it involves birthing AGI. That will create an instant and permanent divide the likes of which will never, ever be cross; you will either be an owner of intelligence indistinguishable from a god, or you will be a mortal. Even the RISK of this happening is laughable that it is being allowed.

We need radically redesigned government, regulation, and public involvement, and we need it yesterday. AGI is a Earth-wide, publicly owned effort, it cannot be relinquished to the owner / slaver class of this planet -- that is madness.

In two posts, you went from ~"it's a travesty that only 5000 people have access to the technology that will soon own the world", to ~"it takes at least three years for the state of the art to run on a box owned by myself in my bedroom".

This is a reason for optimism.

Do you understand what AGI is?

Why are people so willing to trust such a small minority with power like that?

If there is even a 1% chance that they decide to "cut the cord", it is still too high. Once AGI is achieved, there is no coming back. Minutes will be like an eternity, and days, let alone years, will be beyond that.

Trusting such a small number of people with that kind of power is obscene, especially when history has time and time again shown what humans do to things which are no longer useful to them.

There has never been a divide like humans w access to AGI and regular humans before. It will be greater than the difference between a human with a modern cell phone and a carrier pigeon -- the pigeon itself, not a human using it.

My money is on the cord being cut. That means as soon as AGI is achieved, an impenetrable two-tiered human species is created; one with AGI backed intelligence, and the majority, without it, or with a dumbed down version as a transitional cookie until a more final solution can be realized. Once it reaches this stage (without any change to public governance over AGI), it will be too late.

We are already a stone's throw from this reality, if it hasn't already happened.

Why is my money on this outcome? Because that thinking brings about necessary change. We as a people have the power to prevent it 100%, and we ought to, now. Instead of relying on the outlandish chance that a historically malevolent elite suddenly gains benevolence and shares access from the kindness of their hearts to a boundless intelligence, there is a window to force access for all. Everyone having access is autonomously balancing.

Sounds like we need a public option funding and training foundational models and fueling public research which can outpace corporate models in money and brainpower. This could be a thing if governments weren't horrifically corrupted by corporate interests already, or if we could get off our asses and build some sort of decentralized swarm compute network. I do agree, in terms of raw resources (capital) we are far behind and unorganized. In terms of collective brainpower which could be applied if done right... eh, this shit doesn't seem nearly as hard as the ML experts portray it. Either we're being fed a worldwide false reality bubble, or there's a plethora of low-hanging fruit being discovered daily from just a few foundational model finds, and even though the big firms are gonna scoop those up first it's gonna be pretty hard to ever hide that information for long

Of course they're gonna get a few years heads start - which will feel like centuries in AGI time - but either they wipe us all out during that, or we're gonna put the garbage together and make our own AGI too.

Reply on news.ycombinator.com