Chatbot acclimation

- [[chatbot]] If you make your peers respond to each other's statements, an endless loop will occur.
- It is traditional to design chatbots to not respond to chatbot output because they don't like this

• However, this is due to the fact that it is always responding to input - acclimation Just do it.
• For example, if the same content comes in repeatedly, just ignore it.
  • You could have a buffer of the last 100 or something and store it there.
  • Or hold the past day’s worth.
    • The next day, he’d reply “good morning” with another “good morning.”
  • You could put a “sleep” phase to remove buffers.
    • You can learn something there instead of simply deleting it.
• Nor is it to be judged solely on whether the content is exactly the same.
  • Learning
    • subjective filter “boring”
    • Ignore trivial input
    • Basically, any input that doesn’t hit a keyword is “boring.”
  • The other person didn’t respond back, and my output wasn’t very good.
  • It is enough if the output probability of each response, etc., fluctuates daily.
  • Should I gradually dislike the person who says trivial things?
    • If you have a large amount of information, like Twitter for example, you can filter it by the user’s preferences. - subjective filter
    • Some users are selected because it is difficult to read all of them in the first place.
      • Just unfollow the least interesting person of that user and follow the new one.
    • Not sure if it makes sense to filter when it is possible to read the whole thing
• When this happens, even if the same bot is chatting all the time, “they run out of topics and shut each other down” or “things develop differently even though they are the same stimuli as yesterday” will occur.

---

This page is auto-translated from /nishio/チャットボットの馴化 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.