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 to create a digital twin, multiple attempts can be made virtually to see the improvement before applying the capability to the non-digital twin. Andy is reminded of an article that outlines the affinity between the metaverse and digital twin concepts. Nvidia has a concept of this in their Omniverse capability. Another example of a digital twin with a game overlay is the Job Simulator Game. This game is written as a 2050 historical virtual reality environment allowing the player to experience what it was like to have a job in 2020. This fun VR historical reenactment experience is one of the stories that Tobi Lütke discussed in his recent interview with the Acquired team.
Staying on the VR simulation theme, Andy and Michael take a look at the Rats Play Doom game which trains rats in an immersive way to play Doom.
In the last section of the episode, the team takes a look at some metaverse news. Meta has acquired limitless.ai and is shutting down Rewind on the Mac, and is also shifting more investment from the metaverse to AI. Wrapping up the episode, Michael and Andy look at the Nintendo Virtual Boy and Xteink 4.
What poetry would you write to prompt an LLM? Have your bots 🤖 drop our bots 🤖 a line at @gamesatwork_biz (our home for now) and let us know!
These show notes were lovingly hand crafted by a real human, and not by a bot. All rights reserved. That’s our story and we’re sticking to it.
Selected Links
AI
PC Gamer article: Poets are now cybersecurity threats: Researchers used ‘adversarial poetry’ to trick AI into ignoring its safety guard rails and it worked 62% of the time
arXiv paper: Adversarial Poetry as a Universal Single-Turn Jailbreak Mechanism in Large Language Models
Gilbert & Sullivan: Hail Poetry
The Verge article: iFixit’s FixBot helps with repairs ‘the way a master technician would’
iFixit: Introducing FixBot: We Built an AI That Actually Knows How to Fix Things
Digital Twins
ComputerWorld article: Digital twin tech is a double-edged sword
ComputerWorld article: ‘Digital twin’ tech is twice as great as the metaverse
Nvidia Omniverse
acquired.fm AC2 interview: How to Live in Everyone Else’s Future (with Shopify CEO Tobi Lütke)
Games at Work e490: Codename – “Amelia” (for digital twins)
Doom
Reddit post: Open-source VR framework for training rats to play DOOM
Metaverse
9 to 5 Mac article: Rewind Mac app shutting down following Meta’s acquisition of Limitless
WSJ article: Meta Plans to Shift Spending Away From the Metaverse
Retrododo article: Virtual Boy Accessory For Nintendo Switch/Switch 2 Is Available For Pre-Order
My Nintendo Store: Virtual Boy for Nintendo Switch 2/Nintendo Switch
Tech
The Verge article: This tiny magnetic e-reader sticks to the back of your iPhone
Podcast: Play in new window | Download (Duration: 36:13 — 50.4MB) | Embed
Subscribe: Apple Podcasts | Spotify | Amazon Music | Android | Podcast Index | Youtube Music | RSS | More
