@matigo AI (or specifically LLMs) have one flaw: code (i.e. the "guardrails") and data are fed through the same channel. It's almost like we forgot the lesson John Draper showed us...that in-band signaling is NOT SECURE.

@matigo AI (or specifically LLMs) have one flaw: code (i.e. the "guardrails") and data are fed through the same channel. It's almost like we forgot the lesson John Draper showed us…that in-band signaling is NOT SECURE.