The Socratic Method
Using LLMs is what is known as passive reception, essentially the opposite of active inquiry and critical thinking. You ask a question, and you get an answer. Any critical thinking that needs to be done is done after the answer is provided. The Socratic method is an active one. Rather than the ‘learner’ asking a question and receiving an answer, the ‘learner’ asks for information and the response comes back in the form of a question. This creates an active dialogue between the teacher and learner that allows for the dynamic interplay of ideas rather than the passive uni-directional flow of information.
The folks at OpenAI (pvt) like Socrates and have created a Socratic Method model to help students, currently college students, learn more effectively. Here’s how it works:
You start with a topic.
“Teach me about multi-dimensional models”
The model responds by formatting your question into a series of questions that it poses to you. With a small introduction describing how multi-dimensional models are able to discover subtle differences and similarities that are not easily apparent in low-dimensional space, adding the idea that dimensions are like ‘features’ in humans.
Then you get a question like:
“What might you expect ‘features’ to be if we are thinking of people?”
You answer:
“Height, weight, eye color, hair color”
The AI comes back with how good an answer that was and describes a bit more about low-dimension (1 - 3) features, higher dimension (5 – 100’s) features, and how the more dimensions an object has the more similarities (and differences) between them can be discerned. Then it asks the next question:
“What might be the downside to high-dimensional models?”
You respond:
“More dimensions means more comparisons for each item, which means longer compute time for comparisons and more expense”
Again, the model responds by telling you how insightful your answer is, adds a bit of color to the answer and follows with another question. This repeats five or six times as you get a bit deeper into the concept until the AI reaches some sort of learning module break point and the session ends with a relatively simple summary. While we have tried this system, we have yet to go past two consecutive sessions to see how deep the AI will go into the topic without a specific prompt from the user, but on the surface it seems to be a rather ‘lite’ version of learning, like having someone quiz you on Cliff Notes.
Our main objection, although we like any form of learning, is that the model is obsequious and sycophantic, at least with us, which we expect Socrates might not have been. Too many “Excellent answer, you are right on target” and too few “Close, but you are missing an important point”. Perhaps we are at the age where we don’t need constant reassurance that we can absorb complex ideas and concepts and would rather be challenged, which might be different than a college student who feels pressure to excel and justify a six figure tuition or a grant, but in any case learning is learning. With a few refinements the system could improve the interaction between LLM and user and stimulate better understanding of topic related learning. That said, because it is currently free, and extremely easy to use, we expect it will have an impact on educational AI systems that are currently sold to large institutions and organizations. We expect there will be a higher level version at some point that will carry a price tag, but maybe OpenAI will make it open source for the good of all students across the globe…
The folks at OpenAI (pvt) like Socrates and have created a Socratic Method model to help students, currently college students, learn more effectively. Here’s how it works:
You start with a topic.
“Teach me about multi-dimensional models”
The model responds by formatting your question into a series of questions that it poses to you. With a small introduction describing how multi-dimensional models are able to discover subtle differences and similarities that are not easily apparent in low-dimensional space, adding the idea that dimensions are like ‘features’ in humans.
Then you get a question like:
“What might you expect ‘features’ to be if we are thinking of people?”
You answer:
“Height, weight, eye color, hair color”
The AI comes back with how good an answer that was and describes a bit more about low-dimension (1 - 3) features, higher dimension (5 – 100’s) features, and how the more dimensions an object has the more similarities (and differences) between them can be discerned. Then it asks the next question:
“What might be the downside to high-dimensional models?”
You respond:
“More dimensions means more comparisons for each item, which means longer compute time for comparisons and more expense”
Again, the model responds by telling you how insightful your answer is, adds a bit of color to the answer and follows with another question. This repeats five or six times as you get a bit deeper into the concept until the AI reaches some sort of learning module break point and the session ends with a relatively simple summary. While we have tried this system, we have yet to go past two consecutive sessions to see how deep the AI will go into the topic without a specific prompt from the user, but on the surface it seems to be a rather ‘lite’ version of learning, like having someone quiz you on Cliff Notes.
Our main objection, although we like any form of learning, is that the model is obsequious and sycophantic, at least with us, which we expect Socrates might not have been. Too many “Excellent answer, you are right on target” and too few “Close, but you are missing an important point”. Perhaps we are at the age where we don’t need constant reassurance that we can absorb complex ideas and concepts and would rather be challenged, which might be different than a college student who feels pressure to excel and justify a six figure tuition or a grant, but in any case learning is learning. With a few refinements the system could improve the interaction between LLM and user and stimulate better understanding of topic related learning. That said, because it is currently free, and extremely easy to use, we expect it will have an impact on educational AI systems that are currently sold to large institutions and organizations. We expect there will be a higher level version at some point that will carry a price tag, but maybe OpenAI will make it open source for the good of all students across the globe…
RSS Feed