@colinfortuner: Plugin #4 - ShellPlugin Giving chatGPT direct access to my computer 😅 Plugins are really, really cool @OpenAI Getting to the point where I can just create chatgpt plugins to do anything on my computer
 image

@kmizu: I heard that someone said “English is the strongest programming language in the ChatGPT era”, but perhaps the more universal reference is correct: “ChatGPT does not distinguish between natural languages and formal languages. Perhaps the more universal reference is correct: “ChatGPT does not distinguish between natural languages and formal languages, so programming languages can also be interpreted as natural languages and vice versa”. nishio I agree. And “natural language and programming language” are more expressive than “Japanese and English”, and people who can use various paradigms in programming languages, not only procedural types, are more expressive. kmizu Indeed. I remember from the various paradigms, I haven’t tried logical types (e.g. Prolog) yet.

This one looks pretty good, though. kmizu Prolog is also (as expected) strong! They are backtracking properly.

I would also like to do what happens if you mix (embed) N languages in 1 prompt. image image image kmizu Prolog is almost the only (relatively) major language that can use backtracking and unification, so I think it could be quite useful as a query language when querying ChatGPT. I think it could be quite useful as a query language when querying ChatGPT.

It would be good when you want to generate a lot of solutions, but the constraints are messy as logical expressions.

nishio What to think about when you see something like this ・This is a question for GPT 3.5

  • Since you are asking the question from the ChatGPT UI, the temperature is higher and there should be random shaking, but it looks like you didn’t experiment more than once. ・Interpreting an answer as true without verifying that GPT can give a correct answer to this question.

iritec_jp/status/1644842763131961345 using GPT Python” seems to be a good way to improve the efficiency of programming development.

Because compared to 100% understanding of Python, the other languages are like this according to the GPT responses.

・Python: 100%. ・JavaScript: 80 ・Java: 75 ・C++: 70% ・C#: 65 ・Ruby: 60 ・ PHP: 50%
 more image

nishio I had a conversation the other day about why ChatGPT sets the temperature to 0 first and why not 0 by default. The reason why ChatGPT sets the temperature so high is so that “re-generating” will give a different answer. You should press the regenerate button several times with the same question and observe how much the output changes.

  • nishio The reason why they have such a function is probably because “which of the multiple answers the human chose and kept on choosing” can be used as learning data for “which answer the human prefers”. I think the reason why such a function is installed is because it can be used as learning data of “which answer is preferred by humans”.

@tokoroten: in this day and age, anyone doing “verification” or “research” on ChatGPT 3.5 is a sorry person who can’t pay $20 for the tools they use at work. I don’t think it’s a good idea to be a certified person.

nishio Ah, so this is the “AI research” that someone mentioned the other day as “terrible”. I’m going to mute all of them together under “AI research” because it wastes my time to point out the problem when I see it.


This page is auto-translated from /nishio/旄蚘2023-04-09 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.