Everytime I ask an AI to make a statement "better", without further instructions, the result is often a weaker, less precise, more ambiguous, fuzzier version.
It begs the question of why. What is making the model think fuzzier is "better"? Is it because most texts it was trained on were imprecise and fuzzy? Or is it because it is trying to "average" words to the most common denominator?
GM.
GM
GM VP
Because in journalism we are taught to dumb things down to an 8th grade level and the LLMs have picked that up.
Better = more illiterate.
And also, you have to prompt specifically. “Good” and “better” aren’t great writing feedback even to humans.
It always feels in general conversation, particularly in a professional sense, people are vague and just indirect in how things are communicated. Maybe it’s just picking up the general lines of Western conversation. Lacking gumption and direction.
Exactly. Well put.
Widening the bell curve.
GM 🤔
GM
I imagine people who speak imprecisly lake more impricise answers.
If you make an vague request you'll get a vague response.
maybe trained on less balls-y text. trying to please everyone. like a simp. nice guy.
Generalization lowers the chance of wrongness and usefulness.
Embedded fuzzy logic
I have noticed we have to be redundant and incisive on the wording of the prompt. Even to those virtual assistants that should be already tailored for specific tasks.
It's trained on data that makes statements less abrupt, less confrontational, etc., rather than statements that convey confidence and straightforward thought. This is why I fought with it to get me pithy and confident phraseology for marketing copy.
Probably better mean to make everyone happy. Better is actually a fuzzy word if you think of all the meanings of betters in all the heads of humans
