gg2
Anthropic watermark text adulteration criticized
Archived — this story has rotated out of today’s deck. It is kept here in full.
The gist
Anthropic will watermark Claude's AI text to comply with EU rules. Critics say it degrades writing quality, but experts see little impact.
Background
Anthropic announced it will change how its Claude AI model generates text to comply with an EU regulation requiring all AI-generated text to be watermarked starting in December. The watermark will alter the random choices the model makes in phrasing, leaving a detectable pattern. This has sparked criticism from tech blogger John Gruber, who called it a 'perverse adulteration of what it means to write,' while experts like Steven Murdoch say the impact will be negligible.
How it unfolded
- recentlyAnthropic released an update saying it would change how Claude generates prose to comply with EU watermarking rules.
- recentlyTech blogger John Gruber criticized the move, calling it a 'perverse adulteration' and arguing it would constrain Claude's word choices.
- recentlyExperts like Steven Murdoch downplayed the impact, noting the randomness in LLMs is already essential and the change would be unnoticeable.
Who’s saying what
- Critics
- John Gruber argues watermarking is a 'perverse adulteration' that forces Claude to make worse, less precise word choices.
- Experts
- Steven Murdoch says the change 'probably wouldn't have any noticeable impact' because LLMs already rely on randomness.
- Support
- Some users support watermarking as a way to unmask AI-generated content and protect human writing.