Is it possible to come up with a set of questions, that I answer, and then train an LLM model on, in order for it to reliably replicate my personality, as judged by a personality test?
If we think about a use case where I want to create a kind of copy of myself for future generations (hello ego!) then a personality test is unlikely to be the best measure, however for the purpose of an experiment, it is measurable and hopefully consistent.
Perhaps I could come up with a list of relevant questions, and then take a video of myself answering them? This would give not only text, but also the sound of my voice and my appearance and mannerisms when talking. Even better, I could have my partner ask me the question and flesh it out as a conversation, leading to more natural conversational response styles.
It leads to questions of what is personality? Is it taste/preferences? Is it behavior?
I like the idea of doing this as an open-source project, where everyone can contribute questions, and together we build a framework or policy of questions that result in a high match in personality emulation in a GenAI model.
I’m going to build a little App, that asks me questions, records video and audio of me answering and then saves the results in a structured dataset suitable for training an LLM. Sweet.
Leave a Reply