Improving AI Safety with Red Teaming

We had the honor to join many esteemed speakers at AI Day 2023 to talk about improving AI safety with red teaming.

In the talk, we defined AI safety and what it means to red teaming AI models. We also reviewed recently discovered prompt-based attacks, and demonstrated some of them on ChatGPT.

These fascinating topics are new to us. What we knew came from helping AI clients red-team and defend their products and infra.

We're eager to learn more from everyone. Below is our presentation, please let us know in the comments if you have any questions or feedback:

https://drive.google.com/file/d/1hfxDzAGDEpypzOzWJR1tyqac7T65q4gb/view