At some point I should probably write up a thing on Topic Butchering and User Alignment and how it effectively renders any LLM defenseless over enough of an interaction window.

#AI #LLMs #cybersecurity #infosecurity

Today on Adversarial AI Researcher Jeopardy for $500: US, Guyana, UK, Australia, Canada, and New Zealand cybersecurity posture all benefitted from this researcher's classification of emergent property based vulnerabilities in Large Language Model AI systems.

#AI #LLMs #Cybersecurity #Infosecurity #FiveEyes

The recording of the day I spoke at #HOPE_16 at St. John's University in NYC: https://www.youtube.com/live/6mI-8ias7Dw?si=DjOhx2wcvEg2YqOv

#ai #LLMs #llmsecurity

There's a lot of excitement about updates to Gemini. I can't help but mention to folks that Gemini is terrible at safety. Remember the time @Zigula and I hacked it via Fortnite using just voice chat? We called it the Kyber #jailbreak and it produced instructions for napalm production, baking key like pie, and counting cards. #ai #LLMs #cybersecurity

https://youtu.be/4Go4f-RJnBc?si=JDGmjX8jBi9ibkGk

RE: https://mastodon.social/@nixCraft/115311870741815938

This all tracks well with what I know, but I'd be remiss if I didn't point out the epistemic control and firewalling against consequences that AI/LLMs afford governments are almost certainly why the bubble pop will be followed by a massive bailout.

#ai #tech #LLMs

Maybe it's in response to the whole "I hack LLMs" thing that I do for a living, but I've found myself increasingly taking on manual and technical hobbies.

Pesticide free gardening, whittling and carving without powered tools, carpentry, building itty bitty miniature dioramas, etc.

Damn near magnetic attraction to any hard skill that takes time, focus, effort, and can be done without machinery.

#LLM #AI #Hobbies #DIY

For #JukeboxFridayNight I figured I'd pop an old favorite up. Talib Kweli - Get By

https://music.youtube.com/watch?v=9NklQdD_4UM&si=yFHMZS_Wd7h9XQtK