A Controversial Perspective on Prompt Engineering

While many emphasize specific techniques and methodologies as the "best" way to construct prompts, Jared Zoneraich proposes a different approach. Instead, he advocates for treating the prompt as a black box, focusing on the input and output relationship without excessive technicalities.

[zandrrlife]

You can't argue with empirical results though. I don't understand his viewpoint. Also besides high-level prompt to"...swiftsage). Low-level aspects of prompting are under explored. We know the programming code in LM's pre-training data is the main conduit to in-context learning. So why not use a natural language syntax to elicit this behavior even more?

In terms of low-level prompt formatting. Using a programming-style schema with natural language, regardless of prompting technique, boost results consistently. Just a jewel to the community. LM insights pushed me towards the research literature HARD 😂. Huge competitive advantage. A lot of people lack understanding of fundamental research, because it's so new and a lot of key papers are pre-prints 😂. Literally impossible to keep up with the pace of research without using a personal LM...thank you llama. Zuck is my tech OG.