If we decide not to gamble on that outcome, what would you do differently than what is being done now? The EU already approved the AI act, so legislation-wise we're already facing the problem.
The EU AI act - like all laws - only matter to those which are required to follow said law.
Yes, but it's really hard to see a technical solution to this problem, short of having locked down hardware that only runs signed government-approved models and giving unlocked hardware only to research centers. Which is a solution that I don't like.
The "yet" is carrying a lot of weight in that statement. It is now five years since the launch of GPT-2, three years since the launch of GPT-3 and less than 18 months since the launch of ChatGPT. I cannot think of any technology that has improved so much in such a short space of time.
We might hit an inflection point and see that rate of improvement stall, but we might not; we're not really sure where that point might lie, because there's likely to still be a reasonable amount of low-hanging fruit regarding algorithmic and hardware efficiency. If OpenAI and their peers can maintain a reasonable rate of improvement for just a few more years, then we're looking at a truly transformational technology, something like the internet that will have vast repercussions that we can't begin to predict.
The whole LLM thing might be a nothingburger, but how much are we willing to gamble on that outcome?