Molly White: One flaw of the LLMs I've used: they will never give you harsh criticism. While it would be nice to think all my writing is just that good, I know there are no circumstances where someone will ask for feedback and it will say “throw the whole thing out and start again.”

One flaw of the LLMs I've used: they will never give you harsh criticism. While it would be nice to think all my writing is just that good, I know there are no circumstances where someone will ask for feedback and it will say “throw the whole thing out and start again.”

or draw a little vomiting smiley in the margin, as I sometimes did when grading

"Tonstant Weader fwowed up." - Dorothy Parker

If they could recognize concepts that are flawed from the core, they'd delete themselves.

There is, I think, a serious hazard to the customer service mindset of interactive AI and the lack of dignity built into the system. i-am-j.ag/blog/posts/a...

AI needs to be designed with dignity | Joshua Ginsberg, aka jagi-am-j.ag The case for building dignity into our AI and automation assistants

right because they aren't doing any analysis whatsoever and you'd be better off getting your town psychic to proof your writing

Back in the 60s and 70s people used the I Ching and sun sign horoscopes for this sort of writing aid. I can see it functioning as a rubber duck, but a rubber duck that cost several million dollars and thousands of kW-Hr to generate.

could you prompt it to be?

you can definitely get it to be harsher if you tell it to be, but the default setting seems to be "second grader winning a participation trophy"

Interesting and also disappointing.

From what I've seen from people trying to edit with OpenAI, it mostly can only spit out rote writing advice and given the power to rewrite, can only make your writing mediocre.

There may be a silly politeness benchmark they use for "safety", but it's more likely that sincere criticism is just not available on public websites to easily plagiarize

here's output from gemini 1.5 flash after giving it all of my socials.

It's probably because they don't know how to allow them to without also doing a bunch of objectionable and alarming stuff.

A con with confidence!

Some AI person had a suggestion on the Hard Fork podcast about how to get around this but I forget what format the prompt took and haven't tried it out myself.

Post