Re:The Science of Detecting LLM-Generated Text
by userbinator on Sunday March 01, 2026 @04:31AM
(#47205139)
I've never seen the word "delve" show up with such frequency in the pre-AI era, but now it's an overwhelmingly large signal of LLM-generated text, so I'm not sure where that came from. Ditto for vomiting emojis everywhere.
[ Parent
| on HN ]
Re:The Science of Detecting LLM-Generated Text
by TRiG_Ireland on Sunday March 01, 2026 @06:09AM
(#47205678)
I have heard that the human trainers for early LLM models were overwhelmingly from West Africa, so some of the word choices reflect that, including a preference for the word delve. This now means that humans from that part of the world are now frequently unfairly suspected of being AI.
The rise in prevalence is recent, but older than transformer models by a comfortable margin.
https://books.google.com/ngrams/graph?content=delve&year_sta...
Re:The Science of Detecting LLM-Generated Text
by blahaj on Sunday March 01, 2026 @09:08AM
(#47206808)
The models are designed to be these fake positive corpospeak yes man assholes by fake positive asshole corporations.
Mirroring real human text is only the basis of training. Afterwards they get aligned a.k.a. lobotomized.