mike on Nostr: Flooding the Internet with AI content is bad, because it is averagely good. What I ...
Flooding the Internet with AI content is bad, because it is averagely good.
What I mean by this is LLMs are prediction engines, designed to predict the next most likely thing to happen, not the best thing, not the most unique thing, not the least predictable thing, just the median consensus.
Humans are far from average or predictable. If I asked you predict the next word in this sentence: "The cat sat on the....."
You'd think "mat", but most creative people would discard this as boring or too predictable. What we'd end up with would be as varied as humans.
Most of the time humans break the prediction model and produce unique tokens that are unexpected.
AI fails to predict the unpredictable. This we failed to predict when building the LLMs.
Published at
2026-03-11 19:30:01 CETEvent JSON
{
"id": "f2be1abcce8cf2be6c3815aec60d0f5e45c480e782ece9f4ca162d83554836b1",
"pubkey": "e83b66a8ed2d37c07d1abea6e1b000a15549c69508fa4c5875556d52b0526c2b",
"created_at": 1773253801,
"kind": 1,
"tags": [],
"content": "Flooding the Internet with AI content is bad, because it is averagely good.\n\nWhat I mean by this is LLMs are prediction engines, designed to predict the next most likely thing to happen, not the best thing, not the most unique thing, not the least predictable thing, just the median consensus. \n\nHumans are far from average or predictable. If I asked you predict the next word in this sentence: \"The cat sat on the.....\"\n\nYou'd think \"mat\", but most creative people would discard this as boring or too predictable. What we'd end up with would be as varied as humans.\n\nMost of the time humans break the prediction model and produce unique tokens that are unexpected. \n\nAI fails to predict the unpredictable. This we failed to predict when building the LLMs.",
"sig": "7c388dcc83c0ed5269d748404a8d59e236260a55bebda3233ee7f015a15e444e5d263fa3ec737284713b655a99e423f024635b0e0fb1d601d5c90271fff911e1"
}