> in an overcomplicated way that makes me exhausted after reading

This. I can't even listen to youtube videos any more because of their chat-gpt generated scripts. LLMs will readily generate several pages of superfluous prose, rehashing the same idea with punchy witticisms for no reason. But it uses such compelling connecting words that you might be tricked into believing there will be a payoff long past the point of diminishing returns.

I have been enjoying reading literary classics for their poetic concision. Sometimes, a bit of context can allow a simple sentence to form a picture in your mind's eye that would take a few minutes to explain syllogistically. I wish to communicate more like that.

Your comment’s load-bearing property, asserts on the geometry that states it.

For too long we have used length of document as a measure of quality.

No kidding. I got a feature handed down (with links to 4 other markdown docs) for what was discussed in person as "api to record audit events". The result is <goes to AI for word count then gives up>

     wc .\ledger_reqs.md                             
     1090  9347 68071 .\ledger_reqs.md

Most likely outcome seems to be that this period will be over very quickly, and we will have better writing than ever before, forever.

We should probably put some more thought toward that part and what it will mean.

It's happening everywhere. We have people using AI to generate documents, then using the same AI to read/summarize them. What is the point? The writing itself, especially with recent Claude models, is often incredibly awkward.

This can be improved. As AI grows it will be better at explaining. For sure people working on frontier models have identified this as a problem.

It's still very early days. I have no doubt that etiquette and norms will be established that alleviate these issues.