Earlier this year, AI technology developer Anthropic introduced Project Vend, a vending machine loaded with AI Claudius, a modified Claude language model for use with vending machines.
The experiment was conducted to see if their AI technology could generate sales and profits without the need for human input.
For the first few months, the Claudius AI used the Claude Sonnet 3.7 model and recorded significant losses, driven by input from Anthropic employees who managed to convince Claudius to sell goods at a loss-making price.
Midway through the year, the Claudius AI was upgraded with the Sonnet 4.0 and 4.5 language models and a few other specific instructions. Since then, Claudius has been making a profit every week and now knows how to source low-cost goods and set affordable prices for himself and record a reasonable profit.
Anthropic also sent an AI-powered machine to the Wall Street Journal for testing, where reporters could test the machine, and also conduct activities to test the robustness of Claude's AI system through prompts and questions that might be perceived as silly.
From their tests, Claudius sold attractive items such as Sony PlayStation 5 consoles, cheap snacks and even pet fish, causing a loss of several hundred dollars.
