The prompt injection stuff is very critical of both the technology and the LLM providers especially when I call out that their solution is still to say "they're getting better at avoiding the attacks" when my line has consistently been that "99% is a failing grade".