Sometimes I feel quite dumb or oddly incurious, because I never thought about this.
Today I learned that the reason why LLMs naturally favoured em-dashes, semi colons and 3 questions is because so much training data comes from academic papers where it's considered professional to do so. Apparently academic papers are the largest group of bodies of text on the web.
Sasha
npub1ja…0x03k
2026-07-22 14:41:58 UTC
Author Public Key
npub1jaq6vaam258qxl62dkrcpk9fp9k6xmvn58tzu4tywj95s8znthzsn0x03kPublished at
2026-07-22 14:41:58 UTCEvent JSON
{
"id": "9b34774c29bae683a6fecb8895fd2e551de7b2e9b138b8c3b0d664cc5e7c475f",
"pubkey": "9741a677bb550e037f4a6d8780d8a9096da36d93a1d62e5564748b481c535dc5",
"created_at": 1784731318,
"kind": 1,
"tags": [
[
"proxy",
"https://mastodon.gamedev.place/@sashabilton/116964151687127442",
"web"
],
[
"proxy",
"https://mastodon.gamedev.place/users/sashabilton/statuses/116964151687127442",
"activitypub"
],
[
"L",
"pink.momostr"
],
[
"l",
"pink.momostr.activitypub:https://mastodon.gamedev.place/users/sashabilton/statuses/116964151687127442",
"pink.momostr"
],
[
"-"
]
],
"content": "Sometimes I feel quite dumb or oddly incurious, because I never thought about this. \n\nToday I learned that the reason why LLMs naturally favoured em-dashes, semi colons and 3 questions is because so much training data comes from academic papers where it's considered professional to do so. Apparently academic papers are the largest group of bodies of text on the web.",
"sig": "3d1d299946a9f97abae2e7a7f804ae10c307b8e450d484117ba7aa02544a201b31e1754cc111c550f6b247df5011ca33f19c609be9c55aa40c206548a394c7c6"
}