Anthropic’s Text Watermarking Proves AI Companies Do Not Care at All About Writing
AI companies see words as interchangeable and have zero clue how to judge the “quality” of writing.
Anthropic recently announced that upcoming versions of Claude will incorporate watermarks to indicate that the text was AI-generated. However, the company did not reveal how this would be accomplished, leading to speculation on the podcast. A recent blog post provided details, explaining that Anthropic will achieve this by altering the way its AI generates text.
According to Anthropic, no additional information is added to the text, and readers cannot detect any differences between watermarked and unwatermarked text. To facilitate watermarking, Anthropic subtly modifies word choices in AI-generated text, known only to Anthropic and its algorithms. This approach involves using low-stakes word choices, occurring many times throughout a text, to embed a pattern in Claude's responses.
While readers cannot discern the watermark, anyone with a key could potentially detect it by analyzing the sequence of words. Anthropic emphasizes that watermarking does not affect the quality or readability of Claude's output, and internal testing has shown no impact on content, creativity, or readability. Critics argue that this approach devalues writing and treats words as interchangeable.
The company claims that the implementation aligns with the European Union's AI regulations, but the tone of Anthropic's blog post suggests a lack of concern for the craft of writing.
Written by urgent.news from 404 Media's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
