just terrific
I think that’s a very generous use of the word “superintelligent”. They aren’t anything like what I associate with that word anyhow.
I also don’t really think they are knowledge retrieval engines. I use them extensively in my daily work, for example to write emails and generate ideas. But when it comes to facts they are flaky at best. It’s more of a free association game than knowledge retrieval IMO.
I’m not sure I can give a satisfying answer. There are a lot of moving parts here, and a big issue here is definitions which you also touch upon with your reference to Searle.
I agree with the sentiment that there must be some objective measure of reasoning ability. To me, reasoning is more than following logical rules. It’s also about interpreting the intent of the task. The reasoning models are very sensitive to initial conditions and tend to drift when the question is not super precise or if they don’t have sufficient context.
The AI models are in a sense very fragile to the input. Organic intelligence on the other hand is resilient and also heuristic. I don’t have any specific idea for the test, but it should test the ability to solve a very ill-posed problem.
I’m not saying that we can’t ever build a machine that can think. You can do some remarkable things with math. I personally don’t think our brains have baked in gradient descent, and I don’t think neural networks are a lot like brains at all.
The stochastic parrot is a useful vehicle for criticism and I think there is some truth to it. But I also think LMMs display some super impressive emergent features. But I still think they are really far from AGI.
I definitely think that’s remarkable. But I don’t think scoring high on an external measure like a test is enough to prove the ability to reason. For reasoning, the process matters, IMO.
Reasoning models work by Chain-of-Thought which has been shown to provide some false reassurances about their process https://arxiv.org/abs/2305.04388 .
Maybe passing some math test is enough evidence for you but I think it matters what’s inside the box. For me it’s only proved that tests are a poor measure of the ability to reason.
Do you have any expertise on the issue?
I hold a PhD in probabilistic machine learning and advise businesses on how to use AI effectively for a living so yes.
IMHO, there is simply nothing indicating that it’s close. Sure LLMs can do some incredibly clever sounding word-extrapolation, but the current “reasoning models” still don’t actually reason. They are just LLMs with some extra steps.
There is lots of information out there on the topic so I’m not going to write a long justification here. Gary Marcus has some good points if you want to learn more about what the skeptics say.
We’re not even remotely close. The promise of AGI is part of the AI hype machine and taking it seriously is playing into their hands.
Irrelevant at best, harmful at worst 🤷
What AI revolution? All I get is fancy spellcheck and crappy image generation.
It’s hyperbole.