pull down to refresh
351 sats \ 6 replies \ @SimpleStacker 16 Nov \ on: When do you think the young should learn prompt engineering? AI
IMO the bigger issue is not learning how to prompt, it's learning how to evaluate LLM output.
If you can tell whether the output is good or bad, and especially if you can tell why it was good or bad, you can adjust your prompt accordingly.
To know how to evaluate LLM output, you need critical thinking, domain experience, taste, and agency (to know what you are trying to accomplish). LLM does not do those things for you.
Great point! It reminded me of a screenshot I downloaded from a fellow educator’s sharing. She got her students to think about whether the LLM’s responses are appropriate for the Purpose Audience Context defined by their communication task.
reply
The counterpoint, do we evaluate assembly output of the compilers? Maybe it used to be the case for the first compiled programs, but now we just trust the compilers.
reply
We still evaluate the final output of the compiled program though. And a good engineer will know what kind of edge cases to look for as well. I feel like it'll be that way with LLMs.
reply
What does edge case mean?
reply
Edge case in programming means rare cases where the program gives incorrect output. Like a program that gives the correct output for most inputs, but wrong output for a few rare types of inputs.
In the case of LLMs, I would define an edge case as types of things that LLMs are especially prone to get wrong, or types of nuances that they tend to miss, even if they tend to get most things broadly correct...
reply
Thanks for the education
谢谢老师指点
reply