Altogether, the exercise reinforced my previous opinion: LLMs are terrible writers, but decent editorial rubber ducks.
https://www.mollywhite.net/micro/entry/rubber-duck-editing-with-llms
@molly0xfff FYI: I’m getting a “502 Bad Gateway” trying to read that.
For people missing the post it's the first entry here
Came here to add another LLM rubber duck trick - if it gives advise I always ask the opposite question
Should I x
Then why Should I not x
Edited 323d ago
@molly0xfff I'm getting "503 first byte timeout" from "Varnish cache server" on the shared link, but the post page on the site works:
Also getting an error:
Error 503 first byte timeout
first byte timeout
Error 54113
Details: cache-sjc10061-SJC 1718307845 3633622285
Varnish cache server
@molly0xfff Sometimes you really have to push an LLM to *not* rewrite your work every time… Even when reformatting or summarizing it finds a way to be ‘boring’.
@molly0xfff Great tips. To add onto this, I like to tell AI to critique a piece and to be as harsh as possible. Really helps to find weak arguments and areas people will be quick to criticize.
@molly0xfff yeah, i’m always confused when people assert that the primary use for LLMs is writing, imo it’s their worst use. then again, i like writing, so idk, maybe i’m biased
@molly0xfff Just thinking out loud here and with no experience with such LLM feedback, wouldn't one potentially useful prompt be to ask it to read your thing and summarize the key takeaways? Earlier in my life, getting that kind of feedback from a human was helpful. It only had to be a human who was smart, educated to whatever level was necessary, and who knew nothing about my particular topic. 1/2
Edited 322d ago
@joshwayne When the LLM started out by telling me "You've got a compelling narrative and some valuable insights here" I found myself wondering if ChatGPT would ever respond with something like "wow, this draft is real garbage". Probably not.
@molly0xfff definitely not. If you explicitly say "critique this. Be harsh/cruel" it's about as mean as the average HN comment: very analytical, sometimes misses the point, and seems to be arguing for argument's sake
@retri lol yep. in my experiments using it for proofreading i have to repeatedly insist that it not read the whole thing back to me, just point out any errors.