Embark on a trailblazing odyssey for enhanced security in this one-hour podcast featuring Ads Dawson, Senior Security Engineer at Cohere. Explore the challenges and solutions in securing large language models (LLMs) and natural language programming APIs, covering threat modeling, data breach prevention, and defense strategies. Gain insights into the successful "OWASP Top 10 for Large Language Model Applications" project, co-founded by Ads, which identifies key vulnerabilities in the industry. Delve into insider news from the AI Village's 'Hack the Future' LLM Red Teaming event at Defcon31, and learn about the inaugural Generative AI Red Teaming showdown. Discover Ads' extensive experience in application, network infrastructure, and cybersecurity, spanning from startups to large enterprises, with a focus on LLM/AI Security, Web Application Security, and DevSecOps.
Guarding LLM and NLP APIs: A Trailblazing Odyssey for Enhanced Security - Podcast #190
Overview
Syllabus
[] Ads' preferred coffee
[] Takeaways
[] Please like, share, and subscribe to our MLOps channels!
[] Security and vulnerabilities
[] Work at Cohere and OWASP
[] Previous work vs LLMs Companies
[] LLM vulnerabilities
[] Good qualities to combat prompt injection problems
[] Data lineage
[] Red teaming
[] Freakiest LLM vulnerabilities
[] Severe Autonomy Concerns
[] Hallucinations
[] Prompt injection
[] Vector attacks to be recognized
[] LLMs being customed
[] Security changes due to maturity
[] OWASP Top 10 for Large Language Model Applications
[] Gandalf game
[] Prompt injection attack
[] Overlapping security
[] Data poisoning
[] Toxic data for LLMs
[] Wrap up
Taught by
MLOps.community
