They will spend the next ten years drip feeding the latest model and no one will be any wiser.
As I understand it, the idea of AGI is that it will be able to “invent” stuff, that it will be able to piece together information to create something new. Is this not what they are basically selling?
And if so, in what world do they see themselves this ever being available to the public? There is literally zero percent chance AGI will ever be given to the public. The CIA/FBI will stash it in a black box and use it to continue exploiting the world with a few “here and there” releases for society.
I also don’t want to say I could be wrong, because that’s how I understand “AGI” based on what these people market it as. And in that case… OpenAI will keep taking it nice and slow so as long as no model gets close to them based on a pre-determined margin.
> And if so, in what world do they see themselves this ever being available to the public? There is literally zero percent chance AGI will ever be given to the public. The CIA/FBI will stash it in a black box and use it to continue exploiting the world with a few “here and there” releases for society.
If the board actually believes in it's own mission (which is up to debate) then they'd have a hard time arguing for anything else than a open/FOSS release of the mode straight up, because otherwise yeah, that would likely happen.
Would they release an AGI with full Internet coverage? You can un-censor a normal sized model in 2 days, depending on the AGI architecture let's say less than a month until it's telling you how to implement an untraceable harassment campaign against AI alignment researchers worldwide so the next model they release will be able to tell you how to clone smallpox.
As I understand it, the idea of AGI is that it will be able to “invent” stuff, that it will be able to piece together information to create something new. Is this not what they are basically selling?
And if so, in what world do they see themselves this ever being available to the public? There is literally zero percent chance AGI will ever be given to the public. The CIA/FBI will stash it in a black box and use it to continue exploiting the world with a few “here and there” releases for society.
I also don’t want to say I could be wrong, because that’s how I understand “AGI” based on what these people market it as. And in that case… OpenAI will keep taking it nice and slow so as long as no model gets close to them based on a pre-determined margin.
If the board actually believes in it's own mission (which is up to debate) then they'd have a hard time arguing for anything else than a open/FOSS release of the mode straight up, because otherwise yeah, that would likely happen.