Accidental Watermark Leak: Kimi Picks Up Claude's Mark via Training Data
yuxiangw_cs · x · 2026-08-11
A user shared an amusing case of AI data contamination: Alice used Claude to generate a large amount of text and then asked Kimi to paraphrase it to avoid AI detection.
However, Kimi had scraped Alice's data during training, inadvertently inheriting Claude's hidden watermark. This comedic incident highlights the ongoing challenges of watermarking and data provenance in LLM training.
More from Fun
- Game dev: telling Claude to "make it fast" won't get you 99th-percentile optimization — gdechichi · 2026-10-03
- Papers with 'agent' in the title get cited more: AI lit search reads titles too — rajammanabrolu · 2026-10-03
- First Model to Pass the Video Turing Test: Half of Users Thought It Was Human — Puzzleheaded-King584 · 2026-10-03
- Yann LeCun calls Anthropic CEO Dario Amodei 'deluded' and 'crazy' over cybersecurity claims — yogthos · 2026-10-03
- YouTube is filling up with AI slop that merely restates the blog posts it cites — generativist · 2026-10-03
- Every's Dev Day chat with Matthew Berman: budget gone the moment he tried Ultrafast — every · 2026-10-03