3 ms·
Many commenters (and the paper) are thinking about the watermarking in adversarial settings, e.g. detecting students using AI assistance improperly. But I thin
by aaronharnly 3y ago
Many commenters (and the paper) are thinking about the watermarking in adversarial settings, e.g. detecting students using AI assistance improperly.
But I think even simple watermarking probably has value; consider a corporate context in which the corporation itself may want to monitor and know what proportion of the code, content, or work product is AI-generated. In that setting, fairly simple markers would allow at least a rough estimate or indication, although they'd have the converse problem of not necessarily indicating places where humans did some hand-editing of the output.
- auraham 3y agoThere is a story about Elon Musk tracking the source of a leak in Tesla by adding spaces in emails: > We sent what appeared to be identical emails to all, but each was actually coded with either one or two spaces between sentences, forming a binary signature that identified the leaker. https://theintercept.com/2022/12/15/elon-musk-leaks-twitter/ https://theintercept.com/2022/12/15/elon-musk-leaks-twitter/
- JohnFen 3y agoAnd I personally want some method of detecting LLM output to help protect me in my own internet reading. Even a method that is imperfect would be welcome.