I am a philosophy practitioner, meaning I do philosophy consultations to my clients : I question them, show themp problems in their existence, train them to argumenting, conceptualizong, analyzing. I was wondering if it would be possible to teach the AI to become a virtual philosopher, a virtual Socrates. Starting from the question of a client, the virtual philosopher would ask a question to the client, the client would respond and the philosopher would question again and so forth. Each question would be generated by analysing the problems in the previous answer. Several comments would be made on the client’s being according to his way of answering or not answering. I would need to simulate a lot of dialogues to train the IA to perform this. Anyone has an idea of how I could reach this goal ?
Folks have been doing this since its inception. It’s pretty common and very easy. You don’t need any training data to get started, just tell it to be a philosopher. All you have to do is insert yourself or change it into a chatbot:
Yes thanks I tried it and it works rather well, impressive. It cannot refrain from answering whereas I just want it to question though.
You can add STOPs so that it will stop when it tries to answer.
I would provide it with several examples (or even just one to test) of exactly what you want it to do. Complete the conversation, and use a factual source such as actual Socratic debates, but preceded by some sort of prompt where you T-up the topic of the question.
You are Socrates. Engage in a debate with your student based on the Socratic Method.
Student: your reverse engineered topic goes here
Socrates: this is an actual example of a question he asked
You can repeat this process for as many times as needed to get it to perform well. Once it’s performing well, you can insert the actual topic you are interested in the last prompt, and Instruct-GPT will generate the completion.
In general you can use this few shot method (few shot = a few examples) to steer it towards any format. Or to alter its personality with examples from a different philosopher.
thanks a lot for your input
Thaks a lot, very helpful.