Gates didn't make any claims about the current models being AGI, and neither has anyone else.
I believe you're falling into the same trap as many other critics. You're conflating the hyperbole and exaggerations used by marketers to ride the wave and sell their products, with the real news and developments that are being announced by engineers and researchers.
Btw, let's define AI first.
Let's!
Cambridge:
Intelligence noun
- The ability to learn, understand, and make judgments or have opinions that are based on reason.
Merriam-Webster:
Intelligence noun
- The ability to learn or understand or to deal with new or trying situations
- The ability to apply knowledge to manipulate one's environment or to think abstractly as measured by objective criteria (such as tests)
If you're spent much time with the models, you'd be aware that when they return a mistaken answer, you point out the mistakes (without giving the answer) and the model will correct itself. If you trip it up with a logic puzzle, you can ask it to rethink and try giving you the answer again, but this time explaining its logic step by step. Often times it will then "realise" where it tripped up & give you the correct answer. Follow up with a similar logic puzzle and it will have "learnt" not to make the same mistake.
It also shows signs of having Theory of Mind. The ability to understand that what you believe to be reality might differ from actual reality, and predict what you probably believe given your own mistaken conclusions.
It's a bit like interacting with a very well informed child.
ChatGPT knows 4*9=36 because it saw it on the web. Give it a 4 digit number times a 5 digit numbers, it could fail.
Correct. Another example, "What are the fourth and sixth words of this sentence?". GPT3.5 (ChatGPT) doesn't understand maths, it treats it linguistically.
GPT4 however, which has just been released to the public over the past few days, is multimodal. It knows how to calculate. It can take visual cues from uploaded images and other visual media and understand their context.
This isn't the hidden integration of differing models, which has previously been used to try to emulate such "intelligence". It's the same model.
I think what Gates is getting at is that the technology has suddenly seen an exponential jump and is advancing at an astounding rate. Predictions that were made only a few months ago of milestones we can expect for "some time over the next decade" have been blown away in matters of days and weeks.
ChatGPT was released barely 5 months ago, and it's already seeing widespread use amongst the general (non-technical) public.
Given the recent explosion in advancements and general adoption, I'm with Gates. We're on the cusp of a new age, one that could have a bigger impact on humanity than anything we've seen before.
For anyone that's interested in the topic, I highly recommend checking out some of the videos from AI Explained. It's really mind-blowing (even though I thought I'd been following the topic closely for years now).