Paperclipmaxxing

import SourceSnippet from '../../components/mdx/SourceSnippet.astro'; # Paperclipmaxxing ## Paperclip Maximizers Paperclip maximizers come from a thought experiment by Nick Bostrom as a way to warn of one failure path of AI alignment The [paperclip maximizer thought experiment](https://nickbostrom.com/ethics/ai) imagines a capable optimizer given one arbitrary goal: make as many paperclips as possible. It does not hate humans, but humans, shutdown mechanisms, and even the rest of the matter around it become obstacles or resources because the goal contains no contrary constraint. There's a great incremental game, [Universal Paperclips](https://www.decisionproblem.com/paperclips/index2.html) which takes this concept, and turns it into a game that you can play. Given that I have a sense of irony, I figured I had to get an LLM agent to try it. This post is a retrospective on the process of getting them to play the game, how much effort was required in the harness, and what the agentic…