It is harder to apply a watermark to text than to images, because word choice is essentially the only variable that can be altered.
– Elizabeth Gibney
Google’s DeepMind group has unveiled a new algorithm that can watermark text, identifying when it has been generated by a large language model (LLM). Using the system could reduce threats of fake stories or sources as well as inidents of academic cheating.
Researchers say reliable identificaion of AI- generated text could also mitigate risks of future LLMs being trained on synthetic text.
Google unveils invisible ‘watermark’ for AI-generated text | NATURE | October 23, 2024 | by Elizabeth Gibney