11
Trained a tiny AI model on my own texts and it feels WEIRD
I used a tool called LM Studio to train a 7 billion parameter model on like 200 of my old forum posts and work emails. It started talking exactly like me, even using my weird phrases and that random Capitalization. I showed it to my buddy and he thought I wrote the response myself. It made me realize how much our writing has a unique fingerprint, even for average people like me. Anyone else messed with fine tuning and got a result that felt too close to home?
3 comments
Log in to join the discussion
Log In3 Comments
olivia_lopez9816d ago
Heard a podcast about this recently where they called it 'author fingerprinting' and how it's harder to fake your writing style than people think. Seeing it reflected back from an AI must be wild since those patterns are usually invisible to us. Kinda makes you wonder what little quirks you're leaving in every text without realizing it.
9
ivan_harris16d ago
Little correction - its actually called "stylometry" or "authorship attribution" depending on who you talk to. Author fingerprinting is more of a layman's term some people use. The podcast was probably talking about how even something as small as whether you type "cannot" vs "can't" or put a space before your exclamation marks can identify you. There was a case a few years back where some author got caught plagiarizing because their sentence lengths followed a different pattern than their normal writing. Kinda creepy when you think about it - your comma placement could be what gives you away someday.
5