Comment by giancarlostoro
4 hours ago
> Within a day I start getting feedback that makes zero sense. Seemingly no one can get the RPC to work. I had tested this extensively, the complaints made zero sense. One of the complaints includes the actual request being made and the endpoint is entirely wrong. Not a single character typo, a complete fabrication. I ask where this came from and they point me to the documentation they were sent. Every single thing was wrong. The endpoints were wrong, the required parameters were wrong, there were invented features that do not exist. I am a very easy going guy, and I have genuinely never been so furious in my entire life. I am still angry as I write this.
This is really bad, and if you're going to "rewrite" something with AI, proofread it, especially technical documentation. How many hours were burned (and money) over some AI generated goof?
This is another reason I prefer to hand-write my docs over generating AI docs, I don't want documentation that looks pretty but its wrong. It also forces me to re-read all the output and validate it on my end, and then describe it.
> if you're going to "rewrite" something with AI, proofread it, especially technical documentation.
The entire idea of AI, AFAICT, is to avoid the work that is necessary to understand the thing, which would then permit you to proofread (for technical accuracy, not just grammatical well-formed-ness), let alone write it in the first place.
> How many hours were burned (and money) over some AI generated goof?
But you see, the person whose hours were burned was the parent, here, not the person using the AI. The person feeling the pain is not the person who needs to learn the lesson.
(Why you'd take the parent's docs and just run them through a slop-shredder is another question, but I can see the reasons along the lines of "sprucing them up" or "enriching them", etc.)