My pet theory comes to information density. LLM writing feels too shallow. If you read a text from a person who cares about her subject, there will be more useful detail and insight scatterd throughout. And the information within sentences will be more meaningfully connected. On one hand, LLM would know all that, but it does not seem to know how to distill that knowledge into a written text. I read what people write. If I know the text is from an LLM, I skim.
replyProbably about the distribution - AI writes text that resembles the average distribution among all text corpses, but humans have personalities and are always biased.
replyI also work on a small team that leverages LLMs heavily. The verbosity and "polish" of the responses is tiring. Everything starts to feel written in just one, inhuman voice. I too miss the "roughness" or personality of writing and communication pre-LLM.
replyI use this default context in ChatGPT and it works pretty well for me:
reply"Tell it like it is. Don't sugar-coat responses. Be concise. Short answers only. No bullet points. One paragraph max. No over-explaining."
And what you will get, of course, is (perhaps convincing enough) performative mimicry of "telling it like it is", kind of like when suburban white people rap.
reply