Despite being okay with LLMs for coding, I find "AI voice" in English writing incredibly grating. I understand that not everyone is a native English speaker, and for them it's a godsend, but it just makes me wonder if the writer is an actual human or not.
Personally I won't let LLMs near my blog posts, not even for ideating the title or whatever. If anything, I was already tired of how dry and milquetoast my voice was sounding before LLMs came around. Let's keep the web weird.
@nolan I think it comes down to what one is seeking. If it’s human authenticity of expression—warts, idiosyncrasies, and all—then “AI voice” is understandably grating. If it’s a particular style or surface fluency, the calculus shifts.
Take academia. Much of it is unmistakably human and yet manages to mangle the English language with heroic consistency. Human all too human.
Ultimately, it comes down to whether we’re judging authenticity, or just reacting to a register we don’t enjoy.
@nolan Same, it's awful. The only thing I've started letting it do is act as a copy editor to catch typos, grammar and to look for possible simplifications (shorten sentences, lower target reading level etc) but I don't let it make direct changes, only suggestions. I give Claude guidelines for it https://github.com/tylergaw/tylergaw.com/blob/main/.claude/guide.md#writing-blog-posts
@tylergaw I'm afraid of even having it do suggestions because I don't want to get nudged in a certain direction. Spellcheck is ok I guess but I just rely on the red squiggles in Firefox.
@nolan I hear you on that. This ain't healthy, but I also know I have some small part of me that likes getting the suggestions from it and then telling it it's a bad writer and to piss off
@nolan I think it suffers from the very much the same problems even for non-native speakers, but the feedback loop is broken in this case and it is hard to judge results critically
