The world is being quietly rearranged by people who write very long documents.


The title they went with The Safety-Aware Denoiser for Text Diffusion Models Noisy translates that to

Making text-generating AI safer just got cheaper


A new research paper shows how to make text-generating AI models safer without expensive retraining. This means companies can add safety features to their AI for less money and time.
Making large AI models safer usually means retraining them, which costs millions of dollars and takes months. This new method lets developers add safety features to a specific type of AI model on the fly. It could make safety more common in text-generating AI because it is now much cheaper to implement.
Watch for major AI companies to announce they are using similar "inference-time" safety methods for their text diffusion models.

If you insist
Read the original →