Cross-entropy comparison of LLM responses reveals Kimi's similarity to Claude

A new comparison of 22 chatbots tried to answer a juicy question: do some AIs “sound” suspiciously alike? The write-up says each model has its own writing fingerprint, and the eyebrow-raising twist is that Kimi apparently lands surprisingly close to Claude. That was enough to send the comment section straight into detective mode, with readers treating the study less like a paper and more like a reality-show reunion where everyone is side-eyeing everyone else.

The loudest split was between the “of course they’ll all converge” crowd and the “wait, what exactly is being proven here?” skeptics. One camp argued this isn’t shocking at all: if chatbots keep getting better, maybe they naturally drift toward the same “best” answer, the one that’s most useful and shortest. The other camp was much less willing to grab popcorn and call the case closed, asking how you can compare one company’s model so closely if you can’t see its inner math at all. In plain English: are we spotting a real family resemblance, or just similar public behavior?

Then came the snark. The iciest joke of the thread was basically, “stolen data was stolen — shocking!” Meanwhile, another commenter spiraled into full-on moat panic, worrying that if a rival can copy a model’s style from a short window of access, then the supposed protections around these products may be flimsier than anyone wants to admit. So yes, the article is about AI fingerprints — but the comments turned it into a messy, funny fight over cloning, proof, and whether anyone in this industry has a real edge anymore.