Short promotional/introduction to an online course called “Quality and Safety for LLM Applications” (in partnership with WhyLabs). Andrew Ng and instructor Bernice Herdin outline the course purpose: teach common failure modes for LLM-based apps (hallucinations, prompt injections, data leakage of PII, toxic outputs) and provide practical monitoring and mitigation best practices for safer deployment.
Hi, I'm excited to introduce Quality and Safety for LLM applications built in partnership with ylabs. While many developers have found that you can prototype LLM applications quickly, one huge barrier to the practical deployment has been quality and safety. If a company were to release that chatbot or that QA system, would it hallucinate and mislead users or say something inappropriate? Or will the application open up a new security loophole where a user can input a tricky prompt called a prompt input injection that makes the alarm do something bad? This short course teaches what can go wrong as well as best practices for mitigating problems, including prompt injections, hallucinations, leakage of confidential pii like personally identifiable information such as email or government ID numbers, and toxic or other inappropriate outputs. Understanding these issues in your development process would make your LLM based app store safer and higher quality. I'm delighted to introduce the instructor of this course, Bernice Herdin, Senior Data Scientist at ylabs, whom I've already had the pleasure of collaborating with a few times since YLABS is a portfolio company of my team AI Fund.
Thanks Andrew. This course is designed to help you discover and create metrics needed to monitor your LLM systems for both safety and quality issues.
We hope this course will quickly give you and your team an injection of useful tools to make your OM apps user friendly and safe. And in fact, if you take this course now, then it's not just an injection of tools, but a prompt injection of tools.
Andrew well, I hope you enjoy the course.
