5 ms·

His explanation of how one can feel like they have knowledge on a subject after engaging in some edutainment but you realize how shallow your knowledge is once

His explanation of how one can feel like they have knowledge on a subject after engaging in some edutainment but you realize how shallow your knowledge is once you try to discuss it with others is something I find for myself a lot. It is one thing I hope AIs can really help with. It is a common belief that you don't understand something until you can explain it to someone else in simple terms. I've tried this method with AI chat bots with some success. They have two immense advantages over other humans: infinite patience and bias towards understanding over agreement. For complex and abstract concepts, most people just don't want to sit and listen to a 40+ year old dude mansplain them. AIs - no problem. They'll listen forever and never get tired, bored, frustrated, etc. The second advantage is even more important. Most people listen only to wait for their turn to speak. Or they will get caught up on a minor point because they don't agree. That proclivity to disagree can often turn into blindness, erasing any further information past wherever they got hung up. Even if they do ask you questions, often it is some attempt to highlight their disagreements, in a pseudo-socratic method kind of lawyering. AIs have no ego, they don't have stake in the game. They aren't trying to convince you to agree with them. They can just listen and understand, rephrasing and repeating back.

ethanbond 3y ago
Have you tried this with AI with a topic you are well versed in? I am very optimistic AIs will help people feel they have some sort of understanding but far less optimistic they’ll actually help build it. They’ll try to coerce any of your (or the model’s) mistakes into the correct answer out of politeness, and when you make a mistake it has zero model of the model you used to produce such a mistake, so it has no idea where to intervene in your knowledge except at which words hit the keyboard. Idk, I am extremely extremely unimpressed any time I try to quiz it on topics I know. Except for coding, which makes sense since that’s the exercise of pushing as much semantic information into syntactic information as possible.
- zoogeny 3y ago
  It sounds like you are doing something different than what I am suggesting. I am not quizzing the AI looking to determine how well the AI knows the subject. I am using the AI as if it doesn't know the subject at all. I try to explain a subject to the AI as if I am trying to teach a friend who has no knowledge on the subject. Then I am judging how well I can communicate the idea. In many cases as soon as I try to explain the idea I realize that my own knowledge isn't deep. So I am judging myself, not the quality of the AI. An analogy is the Rubber Duck method of program debugging. Most programers have been in the situation where when they try to explain an answer to another programmer the answer will suddenly pop into their head. The value isn't the knowledge in the other programmers head, it is in the act of trying to explain. LLMs are patient and unopinionated and make for a good recipient.
  - mmcnickle 3y ago
    I was intrigued by this idea and decided to give it a go on a subject I have only cursory knowledge of. The conversation is here (https://chat.openai.com/share/50fe7120-a4eb-4c01-80bc-43d8f7f01f02 https://chat.openai.com/share/50fe7120-a4eb-4c01-80bc-43d8f7...) You're correct that it's an exercise in introspection, rather than relying on the AI's own knowledge. It was clear to me when trying to write explanations to its questions where my understanding was starting to get fuzzy and hand wavy. A nice bonus was to get it produce a scorecard of correct, nearly correct and incorrect explanations. I could see these as a good jumping off point for me to do more learning/research. Though I suspect the AI would be less accurate at this in a more niche topic than I chose (refrigeration).
    - zoogeny 3y ago
      That is an interesting conversation. I like that you had the chat bot follow up with questions forcing you to dig deeper. The summary at the end is a good idea as well. I've been asking the chatbot to point out my assumptions and to point out weaknesses or errors in my reasoning. That has been pretty useful in uncovering aspects of my thought that aren't as foundational or coherent/consistent as I expected. It also doesn't require the AI to be an expert in order to give me feedback, it just points out unchallenged assertions and errors in logic. I think this is a potentially powerful use for AIs. They can patiently and politely nitpick and point out subtle errors in our thinking process, providing unbiased critical feedback. That is almost as valuable as having them be perfect oracles that can answer any possible question.
      - mmcnickle 3y ago
        I like that idea of having it probe the reasoning as much as the content, I'll definitely add that to the mix next time I try this.
    - ethanbond 3y ago
      Wow, that is awesome. Kudos to you and @zoogeny's approach. This seems like a much better way to use this educationally than what I was attempting! Thank you for sharing. It helps that refrigeration has always sort of blown my mind, so this was an interesting topic as well.

ethanbond 3y ago

Have you tried this with AI with a topic you are well versed in? I am very optimistic AIs will help people feel they have some sort of understanding but far less optimistic they’ll actually help build it. They’ll try to coerce any of your (or the model’s) mistakes into the correct answer out of politeness, and when you make a mistake it has zero model of the model you used to produce such a mistake, so it has no idea where to intervene in your knowledge except at which words hit the keyboard. Idk, I am extremely extremely unimpressed any time I try to quiz it on topics I know. Except for coding, which makes sense since that’s the exercise of pushing as much semantic information into syntactic information as possible.

zoogeny 3y ago
It sounds like you are doing something different than what I am suggesting. I am not quizzing the AI looking to determine how well the AI knows the subject. I am using the AI as if it doesn't know the subject at all. I try to explain a subject to the AI as if I am trying to teach a friend who has no knowledge on the subject. Then I am judging how well I can communicate the idea. In many cases as soon as I try to explain the idea I realize that my own knowledge isn't deep. So I am judging myself, not the quality of the AI. An analogy is the Rubber Duck method of program debugging. Most programers have been in the situation where when they try to explain an answer to another programmer the answer will suddenly pop into their head. The value isn't the knowledge in the other programmers head, it is in the act of trying to explain. LLMs are patient and unopinionated and make for a good recipient.
- mmcnickle 3y ago
  I was intrigued by this idea and decided to give it a go on a subject I have only cursory knowledge of. The conversation is here (https://chat.openai.com/share/50fe7120-a4eb-4c01-80bc-43d8f7f01f02 https://chat.openai.com/share/50fe7120-a4eb-4c01-80bc-43d8f7...) You're correct that it's an exercise in introspection, rather than relying on the AI's own knowledge. It was clear to me when trying to write explanations to its questions where my understanding was starting to get fuzzy and hand wavy. A nice bonus was to get it produce a scorecard of correct, nearly correct and incorrect explanations. I could see these as a good jumping off point for me to do more learning/research. Though I suspect the AI would be less accurate at this in a more niche topic than I chose (refrigeration).
  - zoogeny 3y ago
    That is an interesting conversation. I like that you had the chat bot follow up with questions forcing you to dig deeper. The summary at the end is a good idea as well. I've been asking the chatbot to point out my assumptions and to point out weaknesses or errors in my reasoning. That has been pretty useful in uncovering aspects of my thought that aren't as foundational or coherent/consistent as I expected. It also doesn't require the AI to be an expert in order to give me feedback, it just points out unchallenged assertions and errors in logic. I think this is a potentially powerful use for AIs. They can patiently and politely nitpick and point out subtle errors in our thinking process, providing unbiased critical feedback. That is almost as valuable as having them be perfect oracles that can answer any possible question.
    - mmcnickle 3y ago
      I like that idea of having it probe the reasoning as much as the content, I'll definitely add that to the mix next time I try this.
  - ethanbond 3y ago
    Wow, that is awesome. Kudos to you and @zoogeny's approach. This seems like a much better way to use this educationally than what I was attempting! Thank you for sharing. It helps that refrigeration has always sort of blown my mind, so this was an interesting topic as well.