Comment by rpigab
2 months ago
I'll buy the AI LLM that finds the prompt that was used based on the output, and replaces posts and emails by their prompt so I can actually understand the intention of the author and not spend my whole afternoon going through all bullet points with highly redundant information.
/s, of course, but not that unrealistic.
Not that /s, really. If you think about it, what a person writing a long-winded e-mail full of redundant text is doing, is the same work LLM is - they have a prompt in their mind, and they're generating text that "sounds nice" out of it.
AI or not, it would be better if they just sent their prompt instead.
If you don't have the original input, how would you determine the prompt that was used to generate the output?
The process is extremely lossy, but guessing a likely prompt should be doable if they used an LLM just to inflate word count, to reverse it, just remove redundancy and noise. Saves you time.