Скачать или смотреть What are AI guardrails? How do they work?

In this video, you'll investigate different approaches to AI guardrails and look at if they address problems of privacy in machine learning/AI. I'm curious to hear what you liked and learned and how you approach threat analysis/modeling and testing in your AI products/systems. Let me know in the comments!

To learn more, feel free to check out the articles and series on memorization in AI models.

Software-based guardrails: https://blog.kjamistan.com/blocking-a...

External algorithmic and internal alignment (i.e. training) guardrails:

Read all articles in the series: https://blog.kjamistan.com/a-deep-div...

And some citations from the video:

Zhang's presentation on avoiding software-based guardrails:    • Quantifying and Understanding Memorization...  

Purple Llama: https://github.com/meta-llama/PurpleL...

Nemo Guardrails: https://github.com/NVIDIA/NeMo-Guardr...