Friday, February 20, 2026
The reason that genuinely transformative AI may not happen in our lifetimes
Kurzweil explained it all decades ago.
Artificial Intelligence is almost completely a matter of brute force in the matter of its training and execution. Quantity over quality.
The problem is that the rate of improvement in processor size and speed may be too slow.
Imagine that fifteen years from now, AI will only be one thousand times better than today.
A thousand times better than what we have now should not be enough to make a qualitative difference.
Subscribe to:
Post Comments (Atom)
That's very true, but the main problem is that, in terms of raw intelligence, there really hasn't been much improvement since the launch of GPT-4 in March 2023.
ReplyDeletehttps://epoch.ai/benchmarks/eci
We've gotten longer context windows, and various schemes where LLMs bounce off each other to refine the response to an input, but GPT-4 at launch was very nearly as capable as today's top models.
Which is a very bad omen for genuinely transformative AI, if you think about it.
A 1000 times better sounds like a lot...
ReplyDelete