There are people who see the same thing and think, “I should be able to do this,” and then there are people who see someone else do it and then say, “I didn’t think I could do that.”

relevance - There must be power. - In keeping with this expression, “[The power that should be able to do it.

nishio Throw a 5 year old Google App Script written 5 years ago at GPT-4 and say “keep it in Python” and make coffee. chaowatson prompt me. nishio What? “Convert the following Google App Script to Python. chaowatson What? Only? Seriously? nishio I don’t understand at all why you can’t come up with such a simple thing, or why you think you “can’t do it” when you hear how to do it. I do it thinking “of course it can be done” and it’s done as expected. If you doubt it, try it yourself.

nishio Programmers who have experience with different programming languages in paradigm have a “paradigm abstraction capability” and using that capability makes LLM seem like a new programming language moment, of course. Then you can guess, “This is what it should be able to do,” and often you’ll be right! - Acquire a model

  • image
    • Usually we acquire model A by abstracting from many concrete experiences, as in 1.
    • If you have acquired abstracted X from multiple models B, C as in 2, you can acquire a new model D with relatively little concrete experience

This page is auto-translated from /nishio/できるはずだvsできるとは思わなかった using DeepL. If you looks something interesting but the auto-translated English is not good enough to understand it, feel free to let me know at @nishio_en. I’m very happy to spread my thought to non-Japanese readers.