Hacker News Reader

Comment by WFHRenaissance | Hacker News Reader

Prompt engineering. The perfection of "programming as verbal intelligence".

I'm working on prompt engineering. I can confirm this is the next big thing that hardly anyone is thinking about.

I'm getting good results from simple approaches to storing and retrieving memories of past conversations and crawled/indexed documents. This is being built at https://mitta.us/

https://openreview.net/pdf?id=Bx-fUfKedZ5

inventive-anteater|> The website discusses an approach for improving the accuracy of the GPT3 language model by providing feedback to the model based on user interactions. The goal is to allow users to interactively teach the model to avoid common mistakes, such as misinterpretations of word meanings. The implementation of this approach is described, and four tasks are used to demonstrate how the model can be substantially improved with user feedback.

I can't help but feel that this will be a fleeting skill.

As humans work towards clearer communication with AI, AI will be working towards clearer understanding of humans.

You could make the same argument for Google searches. In my experience, yes, Google is trying to get better at understanding the (average) user, but average user really makes no efforts to get better at even basic ways to search better. Obviously advanced users try to get better, but both ends require engineer better search engines and better understanding of advanced user interfaces.

In the case of Dall-E, isn't a hack for that to design _another_ ML system that looks at the training data 'descriptions' and infer a correct prompt strcture for the desired result. so the human uses this secondary system as a 'guide' to constructing prompts to navigate them towards the result (image) they want?

I think in some cases you're right. But generally I think the real answer is more along the lines of near-realtime generation of results. That way you can iterate quickly to get closer to what you are looking for, and experiment with different approaches to see which is more aligned with how the AI model thinks about things.

Reply on news.ycombinator.com