Comment by pixl97
6 hours ago
Even a stopped clock is right two times a day, LeCun can't even do that.
>the danger is not intrinsic to the technology
This is why you can't take anything he says any longer at face value. He failed to predict what LLMs can do and now takes the contrary position even when it flies in the face of evidence.
AI safety was a thing before AI even existed. Why, because the outcomes are easily predictable. Give an agent intelligence and bad things can happen in unpredictable manners. Give it even more intelligence and the bad things that can happen only grow worse. This is not some huge new insight. We realized this like, what 70 years ago now?
Now, when we have AI starting to tickle AGI and we're trying to overthrow 70 god damned years of reason and logic on the topic? What the hell.
> Give an agent intelligence and bad things can happen in unpredictable manners
You must mean the opposite, or something quite different from what you wrote.
Give something intelligence and you will have made a remarkable miracle and you will be thanked forever as the Great Benefactor.
Eh, no. I meant exactly what I said.
In our concepts "giving intelligence" and "creating a liability" are directly opposite. You must be attaching an odd idea to 'intelligence'.