Emoji use was stable from 2020-2024, then spiked in 2025. The authors don't attempt to explain it, but I bet AI is to blame. Anyone who has had to clean up AI comments riddled with stupid emojis from their code will understand this.
I wonder if thereis a good explanation why emojis have had such an overproportional impact on LLM output. The em dashes I get, they are quite typical for formal writing, but I have never seen many examples of emoji-littered documents outside of LLM output or chatrooms before.
Maybe I'm just out of touch with some corners of the Internet that this was trained on.