★ Anthropic’s ‘Watermark’ Text Adulteration in Claude Is a Perversion of Writing
It’s unacceptable for a tool to sacrifice an iota of clarity, coherence, meaning, quality, etc. for the purpose of embedding hidden clues within the text to suggest its provenance. The idea that anything other than *my* needs should factor into the generation of text *for me* is patently offensive.
Anthropic has introduced a "watermark" feature for its Claude models, which will embed invisible clues within generated text to assist in detecting AI-generated content. This technique, known as steganography, involves subtly altering the choice of words or tokens during the generation process based on predetermined green and red word lists.
The system does not change the meaning, quality, or readability of Claude's responses, but adds a subtle fingerprint that can later be probabilistically detected. While Anthropic claims this watermarking will not alter the semantics of the text, it effectively does so by subtly changing the wording choices made during generation.
The accuracy of detecting AI-generated content depends on the number of words generated, with larger texts providing more reliable results. However, only Anthropic will have access to the secret key needed to detect these watermark fingerprints.
Written by urgent.news from Daring Fireball's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
