This technology will be as critical as the race for the Nuclear Bomb was back during World War II.
Yes, certainly; no one doubts that. The difference is that in 1945, physicists were, broadly speaking, building a bomb based on a complex formula, and they controlled its detonation. The bomb was powerful and difficult to build, but it remained inert as long as nobody triggered it.
From the very beginnings of LLMs (see Wolfram’s article in particular), mathematicians were already being outstripped by the capabilities of these models. Yes, the results made sense as a consequence of how the system had been programmed, but they were far more sophisticated and relevant than anyone had expected. Faced with the billions upon billions of combinations produced by the ultra-complex networks automatically constructed during training, the quality of these results was, even then, inexplicable.
That’s the difference: with the bomb, we built an object; with AI, we are building intelligence. And that intelligence, which is increasingly surpassing us, is also becoming increasingly difficult to control.