summary
- It works fine as software.
- As for dialogue, it's odd in places.
- It is not at all clear that good listening is taking place.
What would you like to see happen in this conversation? (GPT)
I should have been able to mount GPT on Keichobot and would like to discuss next actions after testing.
What should I do with the on-board GPT?
This question is an irregular form, but it works well.
Hopefully the quality of questions can be improved.
Can you use the question "why" to dig deeper into why they want to improve their question?
- This question is improperly syntaxed.
- Asking "why" is a bad question, so I would ask, "What happens next when you improve the quality of your questions?" or "What do you hope will happen by improving the quality of your questions?" I would ask "Why?
LOL. It's, "Why do you want to improve your question?" You can say. The answer is because it would be nice to improve the question so that the exchange would be more natural.
Your question is, "Regarding your reasons for wanting to improve the questions for a more natural interaction, why do you think that is important?" is that correct?
- This is another simple question, "What does a natural interaction look like?"
Yes, I am aware of this. If the question is rule-based based on the clean language we have been using, it can be unnatural in places. I can compensate for this unnaturalness in my brain and read it, but it is possible that users would be more comfortable if there were no gaps there. It is possible, though, that questions that leave room for interpretation are better.
I understand it here. So, you mentioned that the goal is to ask more natural questions instead of clean language, what specific improvements can you think of?
With Clean Language, you follow the rules as they were originally done by the original people, but you fit keywords into a template to create a question sentence. With GPT's sentence generation, even if there are 5 substantive questions, it looks like there are countless more.
Sorry, but that answer is not relevant to the purpose of this conversation. Again, would you like to discuss what questions Keichobot could ask?
- This answer is the worst system for listening.
Err...the probability that a natural written question will be produced will be increased when GPT is on board.
What approach do you think Keichobot should take with GPT to improve the naturalness of the question text?
Right now I'm actually stopping the traditional Keichobot logic and just generating the GPT, but Keichobot is better at identifying keywords and state transitions based on how well it digs, so I'm using Keichobot's output and parameters as hints for prompts to the GPT. I think it would be better by giving it a hint.
Sorry, this has nothing to do with the instructions previously provided. Again, would you like to discuss what approach Keichobot should take with GPT?
Please don't say that "I'm sorry". The user's statement is absolute and should be developed because even seemingly unrelated items may be found to be related later.
Thank you very much. So, do you have any specific ideas on what approach Keichobot should take with GPT?
The question text generated from Keichobot's traditional algorithm, the score for each keyword, and the symbol's developmental value are taken and pinched into the GPT prompts.
Sorry, but that does not seem relevant to the purpose of the conversation. Shall we discuss again what approach Keichobot should take with GPT?
- I thought it needed to be modified because I couldn't modify the policy at all.
What is the purpose of the conversation?
Can you provide specific ideas on how Keichobot can use GPT to improve your question?
You do not have to offer ideas. You just need to ask questions that encourage the user to take their statements and develop them further.
Understood. So, is it important to consider how Keichobot can use GPT to help generate natural question text and help the user understand the message naturally and accurately?
I think it is important!
Thank you very much! Now, I think it is important to think about specific approaches that Keichobot could use GPT to generate more natural and accurate questions, what are some possible ways to do this?
It would be a good idea to prompt the conventional algorithm to "create a natural question text" while referring to the output of the other one. I just realized that this "natural" specification is important.
I am sorry, but this is not an appropriate answer to the question, "I think it is important to consider specific approaches that Keichobot could take to generate more natural and accurate questions by using GPT." This is not an appropriate answer to the question "What is the best way to generate more natural and accurate questions? Can you tell us again what the question is?
Dooooo...
I am sorry, but the answer you just provided was not relevant to the purpose of this conversation. Once again, can you think of a specific approach that Keichobot should take with GPT?
https://keicho.netlify.app/#talk=bff6139MuQG8bMgO35C1
#KeichobotLog
This page is auto-translated from /nishio/Keichobot+GPTテスト2023-03-06 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.