Key facts
- An unreleased Anthropic AI model has made significant progress on the Riemann hypothesis.
- The AI model tested hundreds of ideas to advance the proof.
- The AI model coordinated sub-agents to advance the proof.
- Anthropic will watermark text generated by its AI models.
- The watermarking feature will be applied to Claude.
- The watermarking is to comply with the EU AI Act's transparency requirements.
- The watermark is designed to travel with the text.
- The watermark may be removed by significant editing.
An unreleased artificial intelligence model developed by Anthropic has achieved substantial progress on the Riemann hypothesis, a significant and long-standing unsolved problem in mathematics. The AI model reportedly tested hundreds of different ideas and coordinated multiple sub-agents in its effort to advance a proof for the hypothesis. This development raises profound questions regarding the role artificial intelligence may play in future scientific discovery and research.
In parallel, Anthropic is introducing a new feature that adds an imperceptible watermark to text generated by its AI models, including its Claude chatbot. This implementation is a direct response to comply with the transparency requirements mandated by the European Union's AI Act. The watermarking feature is designed to be applied at the model level, meaning it is intended to be embedded within the generated text itself and travel with it. However, Anthropic acknowledges that the watermark may be removed or compromised if the text undergoes significant editing or alteration.