Why is there so much AI slop everywhere?.

Why is there so much AI slop everywhere? Here's the maths:

LLMs predict the next token based on probabilities. Without constraints, they're drawing from everything they've trained on - so outputs regress toward the mean. Generic slop is literally the statistical average of human writing. So when people say AI can't be original, they are correct.

Is it inevitably going to be everywhere? I don't think so:

The model generates a probability distribution over its entire vocabulary for each token. Something like a Claude Skill, a Gemini Gem or a detailed prompt anywhere else modifies the conditional probability P(token|context) - the context now includes "avoid leverage and unlock", "use longer flowing sentences" or "be self-deprecating but confident" (crap examples of tuning, but you get the drift). Tokens that violate those instructions get suppressed, tokens that match get amplified. You're not changing the model exactly, but you are reshaping the probability landscape it's sampling from. And that's where the originality can creep back in - you're pulling it away from the average toward something more specific, more you.

This isn't some new technical skill you need to learn. You're simply articulating what you subconsciously already do when you write. The rhythm, the word choices, the things that make your writing yours. You've just never had to write it down before because you were the one holding the pen. Now you've got a ghost writer, you need to brief them. That's all tuning is.

Is this going to still read like AI slop - to the many AI bounty hunters out there who love to point out when AI has been involved in writing? Probably, they can sniff out AI slop from a thousand characters. But I guarantee it won't be as obviously slop as that to which we've become accustomed over the last 18 months...

So keep calm and learn to tune your AI.

#AI #AISlop #AgentFlow