Someone at work today shared a video of Sal Khan showing off Khan Academy's AI 'tutor' by telling everyone that, to paraphrase from memory, that 'we all know that AI can't do math, but what if they can?' and it's been sitting on me all day.
I know enough about LLMs to know that they can't do math -- that they won't even try -- and that means whatever is doing the math in this demonstration is some kind of layer between all the NLP input processing and the text generation layer. It took all day, but it dawned on me that this is, and so many other value props. for these generative models are, a giant Wizard of Oz scam. Everything of value's going on behind the curtain.
All the 'AI' really adds is the ability to crow about AI.
A
