[raddan]

Moreover, there are areas of expertise that are very hard to reach without special help. I have a PhD in computer science. Earning that degree was a major time commitment, and it gave me close contact with one of the few experts in a niche area. I absolutely loved the experience.

But I did not learn everything I wanted to know. And I can’t afford to take a time-out to go to grad school again for a second PhD. On the other hand, an LLM, in some sense, has access to the world’s knowledge. With some patience and persistence, you can mine almost-expert-level instruction from one.

When ChatGPT 3.5 came out, I excitedly tried this for a topic that I WAS an expert in. The results were discouraging. The model’s answers seemed to reflect the general misunderstandings that people had about the topic. More recently though, I had to review a paper in the same area, and I used a late-model ChatGPT to help check my work. It was an eye-opening experience because it was no longer confused. And it found longstanding misunderstandings I had, buttressing its answers against my skepticism with citations to original work. I came away very impressed. This kind of AI “rubber duck programming” is my preferred style of use now. I used it just today to help me learn an area of statistics I have always been fuzzy about. This approach definitely requires some careful prompting, but I am optimistic that AI tutors will one day be a real thing. My only worry is that people lose the ability to understand what makes an answer a good one and why we should care about good answers.

[bonoboTP]

A lot of knowledge in such niches has never been written down. Papers don't contain it. Tacit knowledge, it's in unpublished config files and unpublished experiments that some prof or PhD student remembers, all the things that are not "publishing-worthy" (because not flashy, not novel-sounding enough, or a negative result etc) but crucial for the method to actually work. Most of expertise is like this and this is why spending years in a research group, talking to the researchers, participating in their culture, understanding their thinking over years (or even just months) can be invaluable. There are also many things that people only say after a beer, after you know each other well. Claude and ChatGPT haven't ingested any of this. They ingested the sanitized surface version, the press release / advertisement version. Because papers are basically career marketing and advertisement mostly. The real expertise is guarded and hidden.

This is of course about the PhD level and beyond. For regular university-level education, where the goal is to pass the usual exams, ChatGPT and Claude are more than enough. But for real-world "doing", not (yet).

[fluoridation]

>And it found longstanding misunderstandings I had, buttressing its answers against my skepticism with citations to original work.

Did you check those citations to make sure they actually contained the information they were being cited for? Because I've asked difficult questions too and have had mixed results. I would never trust an LLM as a teacher. They make stuff up all the time, and when you're learning you're in the very worst position to be checking the information you're given.