I agree at least to the extent that the information is not being communicated to the audience. A lot of AI discussions at the moment are so high level that they seem to encompass all of statistics and data science rather than being about specific products, which leads to confusion.
I think Leo has also made clear in recent episodes that he is at least tending to speak positively to avoid dismissal of AI in general. While well-intentioned, I think that is problematic for fairly obvious reasons.
I love it when people think I’m ignorant because they disagree with my opinions!
If we define AI as any kind of machine learning (as opposed to deterministic programming) it’s pretty clear it’s coming whether we like it or not. Whether it will be life-changing remains to be seen, but I expect it will cause at least as much disruption as the Industrial revolution.
Would you have wanted to stop the industrial revolution? Could you have?
Seems to me the prudent thing to do would be to ackowledge that it’s likely to happen and do your best to prepare yourself for it. Kind like climate change.
That’s not how most people define ML. By that logic almost no image recognition is AI. I feel like I’m nitpicking, except it demonstrates this perception that you lump every math technique and application together when they are solving different problems.
This whole conversation can be summarized by this quote:
To give an example, I know of a AI that uses a GAN to look at snap shots of crystal structures and makes predictions about the structure of opaque crystals. This AI uses exactly 0 reddit comments in it’s training data, instead it needs domain specific data.
My frustration is that you have not engaged with this simple reality on any show that I can remember, and your response here also does not engage with that reality.
Isn’t image recognition trained on other images? If so, it is, in fact, machine learning, aka AI.
I agree with you, AI does not require Reddit comments.
Which simple reality are you talking about?
Image recognition is deterministic, which you contrasted with ML.
Even generative ML deterministically learn a probability distribution and then stochastically generates objects from the learned distribution. (Believe me, this is hard to grok for a grad student, much less a lay enthusiast. And if you’re a real statistician, it’s all stochastic!)
That all these AIs are so completely different from each other, the most useful ones are extremely domain specific, and that talking about people “judging AI” is like chastening people for discussing “food” when the discussion is on deep-fried ham sandwiches slathered in jelly.
What I hear is the appeal to useful algorithms as “AI” when LLMs are criticized. When you talk about the next industrial revolution though, you seem to narrowly talk about LLMs. Because if you weren’t talking about LLMs, it wouldn’t make sense… The crystal structure GAN from before exists right now!