homeassistant

I see some human patterns in there.

I took your post and gave it along with the prompt "you're a Lemmy user, reply to this post saying the goal of LLM is to mimic human speech"

Guess which two were written by Gemini and which was written by my brain:

A) That’s the thing, though—LLMs are literally trained to mimic human writing patterns and quirks. Seeing "human patterns" in text doesn't really rule out LLM generation anymore; it just means the model (or the prompting) was good at recreating how people actually talk and write.

B) The core purpose of an LLM is literally to mimic human text patterns, so seeing "human patterns" in a piece doesn't mean it wasn't generated or heavily written by AI. That's just the tool doing its job.

C) Considering LLM are literally made to model human language; you're going to see plenty of human patterns in LLM output. Hell that's their selling point - you can't tell it wasn't made by a person!

33% chance you'll guess right and maybe you still do know for real, and that's great! You're right that there is some "Je ne sai quoi" in the way humans speak but these LLM have gotten scary good at sounding like people. Especially if you don't already know the "voice" of whoever wrote what you're reading. Like certain grammatical tells of sorts.

Edit: I just realized this is even harder because my Gemini has a bunch of my writing to use for training data and defaults to trying to mimic my tone and style.