3 ms·
Ask HN: Is mastering prompt engineering a productive use of my time?
Or will this whole idea that we need to learn exactly how to chat with a computer to get the desired output a fools errand which will become obsolete in the near future?
For most of the last year I thought it was a thing that only hobbiest were doing. But I’m starting to learn how many established companies and their very educated engineering teams are just leveraging OpenAI and its relatives by sending paragraphs of text to them. Paragraphs with instructions they’ve spent massive amounts of time fine tuning.
It would seem to me that this is not the best way to get insights out of LLMs in a desired format. Calling a function where you can break
down the criteria you want it to use for its input and output seems much more efficient.
- skadamat 2y agoLLM's are a fantastic tool, but they're just a tool. They're not perfect and you may not always have an LLM nearby or some dark knowledge / data may never be LLM-indexed. Treat it like a search engine! Search engines eliminated the need for some types of skills (scouring physical books for information, using yellow pages to find the phone number of a business, etc) but they didn't eliminate the need for: - journals & newsletters for curating & opining on said information - new media formats for helping us understand the information - and much more We just moved "up the stack" so to speak. Information finding became a commodity. There will be new sources of precious information and work that may now be even more valuable because LLM's commoditized basic stuff.
- AbstractH24 2y agoThat’s an interesting analogy and makes a lot of sense. But in the long run, wouldn’t it make sense to have one or more layers of obfuscation. Rather than everyone typing prompts, have libraries of functions for specialized purposes. Then different models can be plugged in and out of those functions. Just like I can program without knowing assembly. Maybe this already exists. I don’t know.