Actually it is because it learned from human writing styles.
Perhaps you mean "watermarked" rather than "fingerprinted" which is a rather different concept. No, whether or not the output text has been through a watermarking process, it will still retain the same style and substance, because it learned from human writing styles. Watermarking of LLM text works by being more or less transparent and undetectable.
indeed, a lot of people think watermarking means "these words in this order" but it's more like "given this sequence, choose token B rather than token A" and it's the difference between "yes you are right" and "yes you are correct"... and in a paragraph it might be a series of 5 micro-decisions that culminates to a watermark that is otherwise unclear from a shorter string
As an aside, the ring can be recorded and then used to discover damaged parts. If it sounds wrong, it is damaged.
https://grindosonic.com/guides/acoustic-resonance-testing/
Does this work with human bodies?
Everything's a drum. https://www.youtube.com/watch?v=Q2T7oLbmvx4
Love it, but when comparing the recorded version to the generated version, there are significant differences. Is this something which can be improved?
Interesting. I build tuned steel instruments (steelpans and handpans). It's pretty cool to use this with them.
ai;dr
Thanks for this. I'm afraid I'll have to use it every day
Love it, it's a breath of fresh air compared to the usual vibe coded SaaS that's dominating Show HN these days.
Do you know this isn't vibe coded? (Idk if this is or isn't.) I don't mind vibe coded apps that are still good experiences.
It is vibecoded. You can tell from the title alone, but also the code: https://github.com/fraware/EVERYTHING-RINGS
No sane person would write a README this way (emphasis on sane - PR dept drones aren't).
But the dead giveaway is the branches of that repo.
> explicit quality gates, content-addressed provenance, frozen empirical contracts, synthetic qualification
A fellow Opus 5 user I see
Good lord (see also, the docs dir)
The text is a dead giveaway, this is llm style
"something. something else. some short conclusion."
> Tap start, allow microphone access, then make one clean strike. No upload. No account. Your microphone stays local.
The reason LLMs write like that is because they learned on human written taglines.
right, but with the amount of text LLMs generate and how often they use these patterns it's pathological
No. it's so it can be fingerprinted.
Actually it is because it learned from human writing styles.
Perhaps you mean "watermarked" rather than "fingerprinted" which is a rather different concept. No, whether or not the output text has been through a watermarking process, it will still retain the same style and substance, because it learned from human writing styles. Watermarking of LLM text works by being more or less transparent and undetectable.
indeed, a lot of people think watermarking means "these words in this order" but it's more like "given this sequence, choose token B rather than token A" and it's the difference between "yes you are right" and "yes you are correct"... and in a paragraph it might be a series of 5 micro-decisions that culminates to a watermark that is otherwise unclear from a shorter string