@tao @phonner My Uber driver back from the beach yesterday mentioned that he is preparing to go back for his masters, and is taking linear algebra online right now. He has had trouble understanding the way concepts are presented in class, but has been using ChatGPT to ask about concepts in a way that makes sense for him, as a supplement to the class. He said the professor is too busy to give him this kind of personal attention for the amount of time he wants it (about an hour weekly).

Like everything with AI, this both worries and excites me. We have to be careful that it doesn't become a replacement for more personalized attention when that could be available, and we have to be careful to build tools that are either correct or that help users calibrate to their untrustworthiness and think critically about automatically generated responses. If we can do both of those things, I think it can be great. I still worry there are not strong economic incentives to do those two things; often replacement with inferior automation is unfortunately profitable.

@TaliaRinger @tao I share your concerns and, to an extent, your optimism. (Though, maybe my optimism is just the pragmatism of a classroom teacher who knows that these tools are already being used and are here to stay!)

Having watched this space for many years, a separate concern I have is that the evidence offered as proof-of-concept of tools like these are often atypical. The Uber driver motivated enough to go back to school to study advanced math. The HS student driven enough to search out supplementary help videos online. The professional willing to commit to taking an online course. Sure, AI and Khan Academy and MOOCs can add value for these self-starters, but maybe being motivated is the real key to success. And of course not everyone is so motivated!

And as you point out, these things do tend to gravitate more toward providing profit than providing equity.

@phonner @TaliaRinger @tao While AI is certainly a distinct and promising technology, the exact same things have been said (less plausibly maybe) about every new media technology going back to radios.

I think everyone with an interest in this should read Tyack and Cuban's "Tinkering Toward Utopia."

If people want to read something shorter, they might read an essay I co-wrote for the blog Slate Star Codex a few years ago, particularly the last section. https://slatestarcodex.com/2018/09/04/acc-entry-does-the-education-system-adequately-serve-advanced-students/

@mpershan @phonner @tao Hmm, but technology really has opened up a lot to students who learn differently. My handwriting was so bad in elementary school that I needed occupational therapy for it, but I learned how to type extremely efficiently and that has been wildly amazing for me. And as someone who struggles with details, but does very well at high-level intuition, both calculators and computer-aided proof have done wonders for me.

Neural tools and especially language models gain in flexibility what they lose in guarantees. That combination is both more exciting and more worrying. A calculator program is very simple to implement and to fully understand; a formal proof assistant has a deliberately small logical proof-checking kernel meant to be understood by people. A language model is much less limited in what it can take as input, but correspondingly its outputs come with pretty much no guarantees by default, and the functionality of any given language model is obfuscated (I think that part is surmountable though).

Economically, I think there are few incentives to build correct software, because regulation of the tech industry is so lax. As software complexity grows, this becomes more and more of a problem. And software complexity is growing a lot, so this is a symptom of that.

@TaliaRinger @phonner @tao My sense as an algebra teacher is that scientific calculators have not radically opened up higher level mathematics for kids who struggled with calculation, because the higher-level skills often depend on lower-level fluency.

I'm not 100% sure that this is a useful analogy for LLMs or future AI, but I think it probably is. These tools are great -- and will probably be useful as teaching and learning tools -- but I don't expect this latest technology to radically open up education in big new ways.