2025-02-19

I've had a lot of morning thoughts relating to the level of language philosopher operate in. I came across this clip of Derrida in one of the documentaries about him where he was asked to talk about love. He adamantly refused this request but also asked the interviewers to be more specific about the question. Derrida gave it a second thought and attempted to discuss to himself what is a possible entry point in discussing love, in which he babbled a bit about being and the loving subject.

In the same manner, I think this is how we should approach LLMs. In the process of prompting, in asking a question and giving information of what we want the model to do, it must be also clear to ourselves what we are asking and what we are looking for. In a sense, this type of exchange is efficient communication except that we are not communicating with another subject. It is technically a mirror into ourselves, only returning the same amount of thought we put into our prompts.

Similarly, those who can leverage this nuanced prompting reveals the Oracle in the machine.