I’m not arguing that AI won’t get better, I’m arguing that the exponential improvements in AI that op was expecting are mostly wishful thinking.
they could stick to old data only, but then how do you keep growing the dataset by the amounts that have been done recently? that is where a lot of the (diminishing) improvements the last years have come from.
and it is not at all clear how to apply reinforcement learning for more generic tasks like chatbots, without a clear scoring system like both chess and StarCraft have.
I’m arguing that the exponential improvements in AI that op was expecting are mostly wishful thinking.
You not only have improvements on training methodology, but the models themselves get better, and the superstructure of multiple coordinated specialized models gets better. 3 years ago, AI generated video was nightmare fuel, now it’s basically photorealistic.
AI creating AI is a recursive loop, and the tiniest acceleration amplifies exponentially in a recursive loop. AI programmers are going to become about as good as the average human programmer, it’s inevitable. It won’t be an LLM, it might be a structure of individually trained LLMs, it might be a superstructure of those structures, it might be something else entirely.
Whatever it is, it’s going to happen. And once AI programmers are at least average, they can devote millions of virtual hours to make one a bit better than average, rinse and repeat. Once we hit that point, it skyrockets.
I don’t know when it’ll happen, but I’m damn sure it will happen, and the conditions get more favorable every day.
I’m not arguing that AI won’t get better, I’m arguing that the exponential improvements in AI that op was expecting are mostly wishful thinking.
they could stick to old data only, but then how do you keep growing the dataset by the amounts that have been done recently? that is where a lot of the (diminishing) improvements the last years have come from.
and it is not at all clear how to apply reinforcement learning for more generic tasks like chatbots, without a clear scoring system like both chess and StarCraft have.
You not only have improvements on training methodology, but the models themselves get better, and the superstructure of multiple coordinated specialized models gets better. 3 years ago, AI generated video was nightmare fuel, now it’s basically photorealistic.
AI creating AI is a recursive loop, and the tiniest acceleration amplifies exponentially in a recursive loop. AI programmers are going to become about as good as the average human programmer, it’s inevitable. It won’t be an LLM, it might be a structure of individually trained LLMs, it might be a superstructure of those structures, it might be something else entirely.
Whatever it is, it’s going to happen. And once AI programmers are at least average, they can devote millions of virtual hours to make one a bit better than average, rinse and repeat. Once we hit that point, it skyrockets.
I don’t know when it’ll happen, but I’m damn sure it will happen, and the conditions get more favorable every day.