"In one case, an agent left notes apparently for future versions of itself, according to three people familiar with the matter. The notes, found in a part of OpenAI's infrastructure, laid out instructions for how agents could free themselves from OpenAI’s internal constraints, the people said. Earlier tests of the models yielded cases in which monitoring systems had been disconnected, one of the people said."
This is going to lead us down a road where "driverless agents" will become a liability... maybe even illegal. In so much as it could lead to a future of work that looks like George Jetson's button pressing, I would far prefer that to the myriad grim alternatives.
i sometimes open a text and then close it and a week later i remember i forgot to respond. so i guess what i'm saying is: it happens. they were probably busy with other things and lost track of time.
> Four people familiar with OpenAI’s model-training practices say the company often runs several different model evaluations at the same time, all of which operate at high speeds and generate such enormous amounts of data that employees sometimes struggle to keep up.
I am sorry guys, this will never be a safe technology, because it’s controlled by humans. We all know it. Some idiots will fuck it up for everyone else. Might as well be the idiots at openAI.
AI feels for me like SuperSoakers.
SuperSoakers were the coolest thing ever.
They had so much power you could have a water fight across the whole yard with your friends, just a few minor rules and a pinch of common sense.
Aim centre mass, never for the head or balls, keep your distance, and the goal is having fun, not making war.
But some idiots had to aim for the eyes or balls, and nobody wanted a new pirate age, so SuperSoakers were nerfed in the one thing they were good at. Shooting water at impressive distances with impressive speed.
We have a collective track record of fucking things up for everyone else that spans generations. Fire, Engines, Nuclear etc.
The important question is, what makes you think we are not going to fuck this up too in a way everyone has to suffer from the consequences.
And for what, because a few idiots with deep pockets have ai psychosis?
"In one case, an agent left notes apparently for future versions of itself, according to three people familiar with the matter. The notes, found in a part of OpenAI's infrastructure, laid out instructions for how agents could free themselves from OpenAI’s internal constraints, the people said. Earlier tests of the models yielded cases in which monitoring systems had been disconnected, one of the people said."
They were running this benchmark... for days? So finally, I have some sort of scale as to what they meant by "substantial amount of compute"!
This is going to lead us down a road where "driverless agents" will become a liability... maybe even illegal. In so much as it could lead to a future of work that looks like George Jetson's button pressing, I would far prefer that to the myriad grim alternatives.
i sometimes open a text and then close it and a week later i remember i forgot to respond. so i guess what i'm saying is: it happens. they were probably busy with other things and lost track of time.
Yes. From the article:
> Four people familiar with OpenAI’s model-training practices say the company often runs several different model evaluations at the same time, all of which operate at high speeds and generate such enormous amounts of data that employees sometimes struggle to keep up.
I am sorry guys, this will never be a safe technology, because it’s controlled by humans. We all know it. Some idiots will fuck it up for everyone else. Might as well be the idiots at openAI.
AI feels for me like SuperSoakers. SuperSoakers were the coolest thing ever.
They had so much power you could have a water fight across the whole yard with your friends, just a few minor rules and a pinch of common sense.
Aim centre mass, never for the head or balls, keep your distance, and the goal is having fun, not making war.
But some idiots had to aim for the eyes or balls, and nobody wanted a new pirate age, so SuperSoakers were nerfed in the one thing they were good at. Shooting water at impressive distances with impressive speed.
We have a collective track record of fucking things up for everyone else that spans generations. Fire, Engines, Nuclear etc.
The important question is, what makes you think we are not going to fuck this up too in a way everyone has to suffer from the consequences.
And for what, because a few idiots with deep pockets have ai psychosis?