Keichobot automatic summarization function

2023-03-13 I added a function that automatically returns a summary when the chat logs become too large, so I chatted with them to test it out.

memo

• As for the first summary, it wasn't triggered by a bug, so I fixed the condition in the process.
• The OpenAI API's latency has caused the Heroku side to time out and die many times, making it hard to make it practical.
• I feel it would be beneficial to get a summary and next action back to me at the appropriate time in the chat.

log What would you like to see happen in this conversation? (GPT) We've added a function that automatically returns a summary when the chat logs become too large, so let's chat about it while we test it out! Is it a conversation log or another kind of log? Here is the log of this conversation. Is the conversation log you own the log of this chat? Or is it something else? That question is a touchy one. Is it this chat log you are talking about this time? Is it another kind of log? The original algorithm asks certain types of questions and then does not repeat them. This means that if a subtle question is asked, you can just go through it and it will never come up again. With ChatGPT, however, it will come up again and again. Now, can you give us some more of your thoughts on this "subtle question"? Even human counselors can ask subtle questions. This is because they are agnostic about the inner workings of their clients. ChatGPT repeats itself. This repetition is the most frustrating behavior of the client when using the ChatGPT version of Keichobot. ChatGPT is a natural language processing model developed by OpenAI and used for artificial intelligence dialogue generation. This next one is where you give explanations without asking questions. ChatGPT is a natural language processing model used for artificial intelligence dialogue generation; in the case of Keichobot, this repetition may be considered offensive to the client. Keichobot's Japanese was a bit awkward at times, but his ability as a counselor is better than ChatGPT so far. I hope we can merge them well. How did you hear about ChatGPT? I forgot. ChatGPT is an open source natural language processing model available on the Internet, and Keichobot is one application. If we keep it the way it is, we're going to have to reduce the ChatGPT weighting, right? ChatGPT is an open source natural language processing model available on the Internet, on GitHub and other developer platforms. By the way, we've added a conversation feature to the Scrapbox ChatGPT Connector. You mentioned that you have added a conversation feature to the Scrapbox ChatGPT Connector. 👎ChatGPT Is the Scrapbox ChatGPT Connector a tool that makes ChatGPT available on Scrapbox? Or is it something else? This is a tool that allows ChatGPT to converse with reference to Scrapbox data. Preferable to run on Scrapbox in the future, but not implemented yet. The Scrapbox ChatGPT Connector is a tool that allows ChatGPT to have conversations using Scrapbox data, but what are those conversations? Don't think it's a mixed bag. What kind of bot is "Keichobot"? It's a bot that encourages verbalization by asking questions. What kind of question? The original Keichobot behaves according to a coaching technique called Clean Language. Keichobot is a bot that behaves according to a coaching methodology and can organize the other person's words and approach their inner world through questions. (esp. of men) one's prime (approx. age 30) Keichobot is a question-answering bot that runs on a program and follows a coaching methodology. That's right. What types of questions was Keichobot designed to ask? The questions that take out the human mindscape without distorting it are Can you elaborate a bit more on the concept you described of "taking out the human mindscape without distorting it"? When a question contains an implicit assumption, the respondent is influenced by it and gives a distorted answer. This is very bad because the respondent may misinterpret it as his/her own idea without realizing that it has been distorted by the question. Therefore, it is necessary to have a technique to extract the information without distorting it. Can you give some more specific examples to illustrate the concept you mentioned, that if a question contains implicit assumptions, it will result in a distorted answer? For example, "Is it A or B?" I would answer with A or B, even if it is really C. What is the "C" in regards to? For example, the question of whether it is OK or NG creates the misconception that one has to decide "now" which is which, but in reality, the decision can be postponed. What type of "misunderstanding" did you describe? Can you describe them specifically? A misconception that makes us think there are only two choices when there are really more. Is it important to you to have choices? Do you feel it is of any value to you? It is important to have choices. It is valuable. Too many and you won't be able to decide, but until you do, it's better to have as many as possible. What does it mean to you to have a choice? Can you be more specific? Hmmm, that's a tough question. What is important about "choice?" I guess I feel like I made my own decision. How do you feel about the feeling of "self-determination?" If they succeed after making their own decisions, it makes them feel that they have succeeded and increases their sense of self-affirmation. Conversely, a child who takes the college entrance exam as told by his or her parents and passes the exam may go crazy because he or she does not have a sense of self-affirmation. What kind of self is "me"? The self that proactively chooses to live his or her own life. Is there anything else about yourself that you feel is important? I think it's about being able to say, "I lived life proactively" when you die. It is easy to understand if you consider the opposite extreme. If you think, "I just followed my parents' orders all my life," you will feel like a robot. What exactly is the "bot" you are talking about? An entity that can only move according to orders. What exactly would an "order" look like? Taking away the opportunity for proactive decision-making. What exactly do you mean by "decision making" when you state "taking away the opportunity for proactive decision making"? Decide for yourself what actions you will take. Where do you come from with regard to the "self" you described as "deciding for oneself what actions one performs"? It's there when you realize it's there. Can you tell us a little more about "chi"? When you are aware of your existence, you already exist. Who exactly do you mean by "yourself"? Human beings in general, not referring to any specific person here. What exactly do you mean by "choices"? Its human behavioral options. How do you know what the alternatives are? Sometimes we happen to know, and sometimes we actively take action to know. Back to the story, depending on how the question is asked, it can encourage the discovery of alternatives or, conversely, make them invisible, so a questioner without malicious intent is beneficial What do you think is the best way for the person to find options for action? Basically, I'd prefer more information-gathering action. What types of "actions" can the person take to discover behavioral options? For example, attending events, meeting new people, going to schools, etc. to gain new knowledge, joining new communities, etc. One of the other ways to increase your options for action is to increase your output. This includes writing down your thoughts and talking about them. that's so, isn't it It is desirable for the person to discover more options for action. That's right. By "action," what exactly do you mean? What the specific action is depends on the situation. For example, I often consult with Keichobot about development, and in this case, "implementation" is the action. In this case, "implementation" is the action. There are several methods of implementation, or there are several objectives, and I am not sure which one should be prioritized. 'Consulting Keichobot' is an 'implementation' action, but where are you actually taking that action? The implementation is done in front of a computer. Could you be more specific about "implementation?" For example, what kind of programs are you implementing? Sometimes it depends on the Keichobot itself. How is Keichobot actually developed? I'm running it on heroku right now. How did you gather information to learn about Keichobot?

Summary so far [Test chat log summary

• Test the chat log summary feature.
• The original Keichobot behaves according to clean language.
• With ChatGPT, even subtle questions are returned repeatedly.
• Even human counselors can be asked subtle questions.
• Depending on how the questions are asked, they can encourage the discovery of alternatives.
• It is important to increase action.
• Some actions for implementation, depending on the situation.
• Keichobot has its own implementation.
• It is running on Heroku.

[Next action.

• Make it work on Scrapbox.

The summary function worked. I think I can run it on Heroku. Can you tell us what Scrapbox is all about? A very easy-to-use, knowledge-storing mechanism What do you need to do to know its behavior? ？ What needs to be improved regarding "Keichobot itself?" We are currently in the process of exploring the direction of combining Keichobot and GPT. If we decide to make the transition in earnest, I think it would be better to take this opportunity to rework many parts of the existing algorithm that we don't need. What do you think needs to be improved about Keichobot? I think it would be better to let LLM do the machine learning on its own. Can you elaborate on what you can do? The keywords are scored using machine learning to determine which questions are appropriate. Replace there. What kind of output would you like to see emerge from the "question"? It is good if the user speaks with enthusiasm. We are now doing our own machine learning to determine this enthusiasm. What kind of enthusiasm would you like to see as your desired output? It is desirable to get the user into a state of immersion and a lot of talking. New discoveries can be made from what is said in this way. What kind of enthusiasm would you like to see as your desired output with respect to "The Question?" It would be nice to have a situation where words start pouring out in response to questions asked, and more and more verbalization occurs without the user having to work hard at it. Who exactly do you mean by "users"? Me first, then anyone who is interested and wants to use it. What is the "Keichobot" program? A program that encourages such people to go further and further ahead by asking them questions and encouraging them to verbalize their Next Actions. Can you explain exactly how you would pose the question? Right now I'm throwing the whole mess to LLM, but it's not very good, so I guess I'll have to break it up into steps to execute it. If Keichobot is executed in steps, what are the specific steps? That's not enough experimentation yet, and it's hard to try it while tinkering with a system running on a server, better to create an experimental environment with local scripts. What was the purpose of Keichobot? The effect of encouraging clean language verbalization is very impressive, but it is too expensive to make an appointment with a human for coaching, so it is worthwhile to allow a machine to do it, even if it is somewhat less capable than a human. What exactly do you mean by "verbalization?" To be able to put into words and explain matters that I cannot yet explain well. To begin with, who is Keichobot? It would be digital stationery. The relationship between "Keichobot" and "questions" is that Keichobot poses questions to users, encouraging them to deepen and verbalize their own thoughts. Yes. How do you expect Keichobot to change the way "users" think? If you cannot verbalize your thoughts, you cannot look them over. Verbalization is the first step in being able to critically examine your thoughts and find solutions to the contradictions and problems you are struggling with. It would be nice if "users" could deepen their own thinking and verbalize their ideas so that they can examine them more critically and find solutions to conflicting problems, but what else do you expect? If you can verbalize it, others and your future self can read it. Ideas in your brain disappear quickly. Do you expect anything specific to happen to "you"? I expect realizations to happen. I can tell you that after the verbalization happens, you will have a clearer picture of what is important to you, but do you expect it to lead to any specific actions?" Clarity of what is important changes the priorities of resource allocation, which in turn allocates those resources to actions that were not previously allocated mindshare, time, or money, and they begin to move. What specific "actions" do you take once you realize what is important? Actions are created to achieve what you think is important. We hope that their actions will lead to concrete steps to solve problems and issues. (esp. of men) one's prime (approx. age 30) Do you think that this verbalization can be more useful to others and to your future self than to yourself? Of course I think there is. Is this self the conscious self that exists in the brain? Or is it a physical self? A curious question. Since consciousness alone would not be able to act, it is the physical self in the sense that it is the subject that acts. It could be said that the physical self acts according to a course of action verbalized by the conscious self. We can think of this as a link between verbalization and action, which changes the priorities for resource allocation, allocates resources to previously unallocated actions, and creates movement. That's right. As verbalization leads to action, the priorities for resource allocation will change and resources will be allocated to previously unallocated actions, creating a movement that is eagerly awaited. Yes, ask any questions. Verbalization allows you to clearly understand your thoughts and feelings and thereby grasp the values and goals you hold dear. Question. I think "verbalization" can be very useful to understand one's inner life clearly, but what does it mean to you? If you have a clear understanding of your inner self, you can live better The "self" refers to the self as a being that understands its own inner self and values and is better able to lead a better life. That's a good summary. Do you think there is a connection to Keichobot, a clean language designed to provide choice? Clean Language is not designed to provide choice. Are the "options" drawn from your self-narrative? that's so, isn't it So what do you think is the purpose for which Clean Language was designed? It's about drawing them out. The purpose of "clean language" is to perceive information for the purpose of explaining it. Hmmm, well, yes. You could say that the goal is to make the client perceive it clearly so that he/she can explain his/her feelings and thoughts to others. Maybe the related symbolic modeling is clearer on the idea there. By "others," what kind of people are you referring to? Other than yourself What characteristics does he/she have? You can't know what's in your heart without describing it in words. What do you mean by "the word?" The limits of ChatGPT memory. Summary so far

• The chit-chat evolves from the topic of testing a chatbot with the ability to automatically return a summary of chat logs.
• Asking the right questions can put the inner human being into words.
• The importance of enabling clients to explain their own feelings and thoughts to others in words was raised as an objective to make them perceive them clearly.
• Clean Language is not intended to provide choices, but it does play a role in the related symbolic modeling.
• Keichobot and ChatGPT must be combined, starting with experiments in local environments.
• FUTURE WORK: Development of a combination of Keichobot and ChatGPT will be promoted.

https://keicho.netlify.app/#talk=E1oJgUsPApdiVirNOUmV #KeichobotLog

This page is auto-translated from /nishio/Keichobot自動要約機能 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.