from Keichobot on ChatGPT Worldview after LLM

  • gpt.iconWhat would you like to see happen in this conversation?
  • nishio.iconI want to put into words the worldview that I have dimly seen.
  • gpt.iconSpecifically, what kind of worldview do you envision? Can you tell us about the characteristics and elements of that worldview?
  • nishio.iconHomo sapiens first began to transmit knowledge by voice. Since voice disappears, knowledge could not be passed on unless people were in the same place at the same time, and the ability of the receiver to understand limited the information that could be passed on into the future.
  • gpt.iconI see that you are aware of the limitations of using voice to convey knowledge. What new methods of communication do you have in mind to overcome these limitations? What are the characteristics of such methods?
  • nishio.iconLater, Homo sapiens invented writing. This was more expensive than spoken output, but it made it possible to communicate information without being in the same place at the same time.
  • gpt.iconI see. So, by making it possible to transmit information through the written word, communication has become possible beyond the constraints of time and space. What do you think are the advantages of the written word for you? Also, what do you think are the disadvantages and limitations of information transmission through the written word?
  • nishio.iconThere is the problem of suppression of thought using concepts that are difficult to express in writing, but we will come back to that in a moment. Written information transfer has already reached a point where a single homo sapiens cannot receive all of the information inherited from the past.
  • gpt.iconI see that, because of the accumulation of past information, it has become impossible for one person to receive all the information. How do you intend to deal with this situation? Also, what do you think about the problem of too much information becoming difficult to select and discard?
  • nishio.iconThis is where search engines were born. It reads more sentences than Homo sapiens can read, accumulates knowledge, and returns knowledge in response to Homo sapiensā€™ requests. In its early form, it could only return pointers to sentences containing strings of characters entered by homo sapiens. It is only in the last few years that both input and output have become natural language.
  • gpt.iconI see, with the creation of search engines, it has become easier to obtain necessary information from a large amount of information. What do you think are the advantages of this technology? What do you think are the advantages of this technology and what challenges do you see for natural language processing search engines?
  • nishio.iconI wrote elsewhere about what came to mind in the process of writing this. To summarize, ā€œwhat has not yet been put into wordsā€ cannot be the subject.
  • gpt.iconI see, so if something is not worded, it cannot be targeted by search engines. How do you plan to deal with this? Also, what are your thoughts on methods and approaches for putting the unspoken into words?
  • nishio.iconPersonally, I have visual thinking before linguistic turn of mind. In terms of symbolic modeling, there is a space of symbols before language, and we experience it with our senses. I think it is the same thing.
  • gpt.iconI see, so for you, visual thinking precedes linguistic thinking. Also, in terms of symbolic modeling, there is a space of symbols before language, where what is experienced by the five senses is verbalized. What is your opinion on this idea that what is experienced by the five senses exists before language? Also, what are your thoughts on visual thinking and the difficulty of putting into words what is experienced by the five senses?
  • nishio.iconIt requires the power to consciously cut and connect the link between the abstract symbols of words and the concrete senses. In the history of mankind, however, letterpress printing brought about a period in which the cost of reproducing text fell while the cost of reproducing figures remained high. This led to the de facto adoption of a form of expression that ā€œcommunicates primarily through writing. This has resulted in the ā€œHabit of thinking by lettersā€.
  • gpt.iconI see that, while it is necessary to consciously switch the link between words and concrete sensations, the lower cost of reproducing letters through letterpress printing has led to the habit of thinking primarily through letters. What is your solution to this situation? Also, what do you see as the best way to balance the concrete senses with the written word?
  • nishio.iconI think we need to consciously connect the written word to concrete sensations and experiences, not just the written word as it is. Because for many people that is unbalanced and lacking.
  • gpt.iconI see, so you are saying that it is important to connect the text to concrete sensations and experiences. Do you think that for many people, this balance is lacking, and that in order to remedy this, it is necessary to create situations that facilitate access to concrete sensations and experiences? And what do you think are some of the ways in which we can connect concrete sensations and experiences with the written word?
  • nishio.iconI canā€™t start with abstract words, so I guess I would first train myself to express my experience in words. Then, Iā€™ll take the ā€œtentative verbalizationā€ and brush it up by reviewing it over time.
  • gpt.iconI see, so you are saying that training to express oneā€™s own experiences in words is an important first step in connecting words with concrete sensations and experiences. Do you also believe that by reviewing and brushing up your tentative verbalization over time, you can make it more accurate and expressive? Or do you believe there is another way?
  • nishio.iconAt the moment you first verbalize it, you assume that the expression is sufficient. After a period of time, the working memory disappears, so they reread it and realize, ā€œThis is difficult to understand, it is inadequate. It is necessary to develop the ability to notice insufficiency by repeating this process.
  • gpt.iconI see, so you are saying that it is necessary to reread the expression after some time has passed, because at the moment of first verbalizing it, we may assume that the expression is sufficient. In doing so, do you think that it is necessary to develop the ability to notice insufficiencies by repeating the process of reading it back and realizing that it is inadequate? Also, what do you think are some of the ways to foster the ability to notice inadequacy?
  • nishio.iconItā€™s good to have people read it. In that sense, it is also good for language models to show and communicate. Itā€™s the same way I used to practice writing Japanese so that I could go back and forth between Japanese-English and English-Japanese with machine translation. Awareness.
  • gpt.iconI see, so you are saying that having others read and interacting with the language model by showing it to them is a good way to develop the ability to notice inadequacies. You are also saying that you can notice inadequacies in your writing by translating your own writing back into other languages, as in machine translation. Do you believe that by utilizing these methods, you can acquire more accurate and expressive language?

This page is auto-translated from /nishio/LLMä»„é™ć®äø–ē•Œč¦³ 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. Iā€™m very happy to spread my thought to non-Japanese readers.