e535 — The Poetry of DOOM

Photo by David Klein on Unsplash Published 15 December 2025 e535 with Michael M and Andy – adversarial poetry to jailbreak LLMs, iFixit’s FixBot, power of digital twins, putting the breaks on Rewind, Nintendo Virtual Boy and a whole lot more. Michael M and Andy start things off with a most intriguing concept – adversarial poetry.  By using ‘memetic language’, researchers formulated prompts with imagery and metaphor instead of direct operational phrasing to trick LLMs into providing unsafe responses.  Michael makes the point that AI prompts are becoming more and more like spells or incantations.  See the show notes below for a link to the paper for any budding AI poet laureate wannabes.  Perhaps Jabberwocky can be used in a snicker snack way.   Switching to another AI use case, Andy and Michael discuss the iFixit FixBot.  The FixBot provides expert advice and guidance for repairs, by talking to the human who likely needs both hands to effect the repair.  Next up are a couple of stories on digital twins, and how they leverage game technology.  By taking sufficient data points