Is GPT-4 intelligent? If you can overcome the terrible accent of the presenter (quite similar to mine unfortunately!), the answer seems to be yes!
Worse, even if you decide arbitrarily that it is not 'yes' because it doesn't meet some of the elements of intelligence, such as planning for example,, the answer is still 'yes' down the road because the potential improvements to the system are enough to solve any current problem.
I agree that this conclusion is correct and that consequently GPT-4 and its descendants represent a real danger. Worse again, the descendants are not GPT-5 but just some add-on links to tools which magically improve it to reach perfection. And here, we are not talking about the future but the present!
More than a moratorium on AI research, what we need is some very deep thinking about AI, what we want to achieve and the risks. But is our society capable of this or is it beyond our ability as Eliezer Yudkowsky believes?
No comments:
Post a Comment