OpenAI could deploy a watermark to identify text written by ChatGPT

In this video , Scott Aaronson, a theoretical computer scientist at the University of Texas at Austin who spent two years working with OpenAI on AI safety, mentions a watermarking technique designed to make AI-generated text statistically detectable.

More from Singularity

LLMs do fine on ARC-AGI-3 if they are allowed to search over game logs

Singularity·5

Figure's First Full HQ Tour: From the Lab to the Factory Floor

Singularity·6

Figure AI’s CEO shows the Figure 03 humanoid robot’s feet, designed to allow the robot to charge wirelessly

Singularity·10

Scaling the physical Singularity: Autonomous 18-wheelers are no longer a 'pilot project'

Singularity·7

ARC-AGI-3 Update (GPT-5.5 High and Opus4.7)

Singularity·3

Anthropic Won't Let You Use Their Best Model. Prediction Markets Are Trying Anyway.

Singularity·3