zlacker

>>giulio+(OP)
"Recent large-scale upticks in the use of words like “delve” and “intricate” in certain fields, especially education and academic writing, are attributed to the widespread introduction of LLMs with a chat function, like ChatGPT, that overuses those buzzwords."

OK, but please don't do what pg did a year or so ago and dismiss anyone who wrote "delve" as AI writing. I've been using "delve" in speech for 15+ years. It's just a question where and how one learns their English.

>>milanc+p6
In the "opinion" of ChatGPT, my style of writing is "academic". I'm not exactly sure why. Perhaps I draw from a vocabulary or turns of phrase that aren't necessarily characteristic of colloquial speech among native speakers. Technically, English wasn't my first language, so perhaps this is something like the case with RP English in Britain. Only foreigners speak it, so if you speak RP, then you aren't a native Brit.

In any case, it's possible to misuse, abuse, or overuse words like "delve", but to think that the the mere use of "delve" screams "AI-generated"...well, there are some dark tunnels that perhaps such people should delve less into.

>>lo_zam+pg
> In the "opinion" of ChatGPT, my style of writing is "academic".

It may simply be glazing. If you ask it to estimate your IQ (if it complies), it will likely say >130 regardless of what you actually wrote. RLHF taught it that users like being praised.

>>bonobo+Lg
And, if you want to have some fun, you could give it your writing sample - but say that it's from a random blog post you found online. See what it tells you on that.

It really is a shame that an average user loves being glazed so much. Professional RLHF evaluators are a bit better about this kind of thing, but the moment you begin to funnel in-the-wild thumbs-up/thumbs-down feedback from the real users into your training pipeline is the moment you invite disaster.

By now, all major AI models are affected by this "sycophancy disease" to a noticeable degree. And OpenAI appears to have rolled back some of the anti-sycophancy features in GPT-5 after 4o users started experiencing "sycophancy withdrawal".

>>ACCoun+Uj
Not everyone appreciates having his speech characterized as "academic" - in certain circles, it's viewed rather poorly - so I'm not convinced of the glazing hypothesis.

ChatGPT certainly makes distinctions. If I give it a blog post written by a philosophy professor, I get "formal, academic, and analytical". If I feed it an article from The Register, I get "informal and conversational". The justifications it gives are accurate.

"Academic" may simply mean that your writing is best characterized as an example of clearly written prose with an expository flavor, and devoid of regional and working class slang as well as any colloquialisms. Which, again, points to my RP comparison.

>>lo_zam+RH1
Does an average user appreciate this?

Do you?

The first question matters because frying an AI with RL on user feedback means that the preferences of an average user matter a lot to it.

The second question matters because any LLM is incredibly good at squeezing all the little bits of information out of context data. And the data you just gave it was a sample of your writing. Give enough data like that to a sufficiently capable AI and it'll see into your soul.