AI Model Safeguards and Red Teaming for Security

9min Snip

00:00

Play full episode

Summary

Transcript

Episode notes

The chapter discusses the challenges posed by large language models (LLMs) in AI, such as hallucination and prompt injection attacks, emphasizing the need for safeguards like grounding filters and prompt shields. It highlights ongoing AI research to improve models and the importance of training techniques and advancements in AI technology. The chapter also explores the role of a red team at Microsoft in continuously testing AI models for vulnerabilities and developing techniques like the crescendo attack to ensure the safety and security of AI products.

Mark Russinovich, Eric Boyd & Neha Batra join us to discuss the state of AI for Microsoft and OpenAI at Microsoft Build 2024. It’s safe to say that Microsoft is all-in on AI.

Leave us a comment

Changelog++ members save 14 minutes on this episode because they made the ads disappear. Join today!

Sponsors:

Cronitor – Cronitor helps you understand your cron jobs. Capture the status, metrics, and output from every cron job and background process. Name and organize each job, and ensure the right people are alerted when something goes wrong.
1Password – Build securely with 1Password - 1Password simplifies how you securely use, manage, and integrate developer credentials. Manage SSH keys and sign Git commits. Access secrets stored in 1Password. Automate administrative tasks. Integrate with third-party tools. Also, check out our INFRASTRUCTURE.md file for more details on how we do secrets with 1Password.
Neon – The fully managed serverless Postgres with a generous free tier. Neon separates storage and compute to offer autoscaling, branching, and bottomless storage.
Fly.io – The home of Changelog.com — Deploy your apps and databases close to your users. In minutes you can run your Ruby, Go, Node, Deno, Python, or Elixir app (and databases!) all over the world. No ops required. Learn more at fly.io/changelog and check out the speedrun in their docs.