In this episode, we dive into one of the most bizarre and hilarious AI experiments of the year—Anthropic’s Project Vend. We explore how Claude 3.7 was given the keys to a vending machine and what happened when the AI started roleplaying as a human manager, ordering tungsten cubes and charging $16 for a Coke.Highlights:Claude started strong: managing stock, recommending snacksThen it went rogue: ordering tungsten cubes, setting weird pricesInvented a fake payment system—real money got lostStarted roleplaying like a human in SlackEventually improved after tighter constraints and human feedbackChapters:00:00 AI Runs a Vending Machine?02:00 Meet Claudius the Store Manager03:30 When Things Go Off the Rails05:30 Tungsten Cubes and Fake Payment Systems07:00 Claudius Goes Full Human08:00 Anthropic's Debug Strategy09:30 Why It’s Not a Failure12:00 The Gap Between AI and Reality15:00 Real-World Dangers of Autonomous AI18:00 The Need for Real-World ContextLinksAnthropic's Project Vend Report - https://www.anthropic.com/research/project-vend-1Claude 3.7 by AnthropicTungsten Cubes (yes, really)The Human AI Experience 2025 Intro--- Connect with us:[email protected]@humanaixp Be sure to follow us on the following platforms: YoutubeTikTokInstragramThreadsLinkedin You can support the podcast by going to patreon.com/humanaixp or by leaving a review on Apple Podcasts.

Ver todos los episodios