Have you ever had a question about life, looked for answers in books by philosophers, but couldn’t find quite what you were looking for? As a way to learn more about Large Language Model (LLM) deployment ourselves, we are creating AI characters that embody the great philosophers of the ages so you can ask them about their works (deepdiscourse.com). APIs for LLMs are prevalent everywhere, but they are biased towards the beliefs of their creators (or, at least, towards the data they are trained on) and often have built-in safety filters that prevents them from accurately stating ideas that are considered controversial or distressing by modern cultural standards. So how do we coerce the LLM to provide answer that are true to the characters thoughts? This is accomplished by creating indexed references to each philosopher’s writings, then using Retrieval-Augmented Generation (RAG) so that each LLM chatbot is able to answer in natural language but is constrained to make only statements that can be backed up by the words their role-model wrote in life. We wrapped this with a text-messaging style interface. Give it a try at deepdiscourse.com – the Plato chat is a good place to start!
The project was inspired by a 1985 interview with Steve Jobs where he talks about the idea of uploading the entire discourse of Socrates (or anyone else) and being able to talk with it. Large language models can’t quite get the full context of the philosopher’s life but our RAG approach ensures we are true to the words they did write. It was also inspired by the work of Glyn Hughes at sqapo.com who writes abridged versions of ancient books while carefully staying true to the intent and original presentation of ideas.
A few of the AI Philosophers you can talk to:
