Подтвердите e-mail

Для публикаций, комментариев, реакций и сообщений подтвердите адрес.

←

Публикация

dudievins

btw, the people pushing this idea seem to have no clue about many neurodivergent people and their habits: we HATE when we don't control the process and can't switch between tasks JUST BECAUSE OF OUR MOOD SWINGS when they removed the parallel paths in Duolingo, it just became SLOW AND BORING

Carl T. Bergstrom

I have a number of colleagues who think that LLMs will be very useful as individualized tutors. One of their arguments is that unlike previous generations of AI tutors, LLMs have the full context window of the conversation and that’s better understand what a student knows and doesn’t know.

Обсуждение

Ответы

dudievins

it's also ridiculously unsustainable, you know: I wonder how much memory/power a LLM needs to remember all this stuff about a single user to be efficient and btw, I've always been VERY sceptical about the existing evaluations and benchmarks bsky.app/profile/carlbergstrom.com/post/3m4fsadnlbs2e

benchmarksbsky.appCarl T. Bergstrom

But if you ask the same question in a new chat window, it’ll get the right answer straight off the bat. That that’s a big problem because I suspect that is how performance is being evaluated much of the time.