Shreya Rajpal on Guardrails for Large Language Models

Live from the venue of the QCon San Francisco Conference, we are talking with Shreya Rajpal, CEO and Cofounder of Guardrails AI. In this podcast, Shreya shares her insights on building guardrails for large language model (LLM) applications. Rajpal discusses how Guardrails AI assesses the reliability and safety of LLM applications, ensuring any input sent to the model is functionally correct and providing a framework for developers to create their own custom validators.
Read a transcript of this interview: www.bit.ly/47t...
Subscribe to the Software Architects’ Newsletter for your monthly guide to the essential news and experience from industry peers on emerging patterns and technologies:
www.infoq.com/...
Upcoming Events:
QCon London (April 8-10, 2024)
Discover new ideas and insights from senior practitioners driving change and innovation in software development.
qconlondon.com/
InfoQ Dev Summit Boston (June 24-25, 2024)
Actionable insights on today’s critical dev priorities.
devsummit.info...
QCon San Francisco (November 18-22, 2024)
Get practical inspiration and best practices on emerging software trends directly from senior software developers at early adopter companies.
qconsf.com/
The InfoQ Podcasts:
Weekly inspiration to drive innovation and build great teams from senior software leaders. Listen to all our podcasts and read interview transcripts:
The InfoQ Podcast www.infoq.com/...
Engineering Culture Podcast by InfoQ www.infoq.com/...
Generally AI Podcast www.infoq.com/generally-ai-podcast/
Follow InfoQ:
Mastodon: techhub.social...
Twitter: InfoQ
LinkedIn: www.linkedin.com/company/infoq
Facebook: bit.ly/2jmlyG8
Instagram: @infoqdotcom
RUclips: / infoq
Write for InfoQ:
Learn and share the changes and innovations in professional software development.
Join a community of experts.
Increase your visibility.
Grow your career.
www.infoq.com/...

[@iAkashPaul]

For other watchers, few more points to look out for-
Users asking to do a valid task but in a different context. Ex. Write me a python code for doing X valid task
Users asking compound questions.
___
Attempting to constrain the output vocabulary & limiting user input size are ways to reduce the surface of attacks. Query decomposition & then matching it with existing or safe constructs could also be another solution