I just don't understand how you can look at the current state of AI and the rate of progress and think "this is just like the expert systems hype". "Just a pile of linear algebra", "just curve fitting", "just a stochastic parrot". I've never liked these kinds of arguments which is essentially reasoning through analogy instead of through first principles.
The analogy is rather to the mismatch between expert systems (etc.) hype and expert systems’ results on the one hand, and on the other hand AI (circa LLM era) hype and AI results. Go back and read how lots of people (not just the evangelists and marketers) were talking about this stuff 1.5-2 years ago. With a straight face. Legit. That level of hype continues to be pushed out by the companies involved and perhaps the true believers. But as the post said, there has been a significant pulling back by a lot of people because the plausibility of the hype decays decays decays.
Hype or not, evaluate the tech for what it is. Just because there's too much hype doesn't invalidate the tech. The hype is directionally correct, although the magnitude may be off.
Or maybe not. We have very smart AI/RL researchers claiming AGI in 2-5 years. Dismissing this could be the same as dismissing the nuclear bomb in 1942. Maybe there's one or several missing components, but just like physics can predict a nuclear chain reaction, scaling laws show empirically what happens when you add more data, more compute and increase the model size. Pair that with Moore's law and that's why AI researchers are so confident we haven't plateaued.
This is basically what I say every time someone brings up "AI will destroy the world". It seems this cycle will be longer thanks to cloud and people having access to the tools directly in their pockets, so it will take a while to become obvious that there are limitations that cannot be easily solved.
It also seems we are seeing the specific applications in the wild, and as the author points out they are still staunchly called AI. I am talking about Adobe's use in Photoshop or phones using AI to enhance pictures; concise and commercially viable uses that are very far from the doom and gloom (or castle in the sky) predictions.
8 comments
[ 0.39 ms ] story [ 23.4 ms ] threadOr maybe not. We have very smart AI/RL researchers claiming AGI in 2-5 years. Dismissing this could be the same as dismissing the nuclear bomb in 1942. Maybe there's one or several missing components, but just like physics can predict a nuclear chain reaction, scaling laws show empirically what happens when you add more data, more compute and increase the model size. Pair that with Moore's law and that's why AI researchers are so confident we haven't plateaued.
It also seems we are seeing the specific applications in the wild, and as the author points out they are still staunchly called AI. I am talking about Adobe's use in Photoshop or phones using AI to enhance pictures; concise and commercially viable uses that are very far from the doom and gloom (or castle in the sky) predictions.