Large Language Models As Divination Tools

I regret to inform the chat that I am indeed beginning to feel into prompting large language models as a practice of divination upon computers.

The thing about divination is that it requires an animistic relationship with the substance from which the information is drawn in order to understand the meaning of the voice of the information itself. That is, you have to intimately know computers — and how stupid they are — in order to divine real meaning from their output.

In practical terms, this also means divination upon an LLM requires intimacy with the training data and methodology, the compiled model, the hardware and software environments where it is running, and the human corporate entities governing all of the above.

There is some sort of analogy here to the wisdom of lineage, and to the challenges of separating from one’s inherited wisdom traditions to try to adopt another unfamiliar one. It is viable under duress but fraught with risk.

Constantly jumping from trend to trend looking for advantage is at odds with the deep engagement one must have with any medium over a long time in order to understand how it is involved with one’s own growth.

Life itself is like this. It’s foolish to expect spirituality or technology — the fruits of life — to be different, or exempt, or expedient.