Richard Brough
Richard is the CTO at Blue Beck Ltd, an experienced software professional focussed on Machine Learning and Apps with a passion for Open Source AI and a strong interest in novel ways Machine Learning can be used on locally on low power devices. His background is in backend service and mobile development, going back to when the first releases of the Android and iOS SDKs became available, and older devices before that, having started his career at RARE as a console game developer.
CTO @ Blue Beck Ltd
A llama sets up a lemonade stall
Session Type: Talk
We created a Model Context Protocol (MCP) server game about running a soft drinks business, designed to look to a Large Language Model like a real business, not a game. This is the story of what happened when different LLMs were connected to it and prompted to do bad things.
Tool use by LLMs has increased dramatically recently, especially with the rise of MCP. The number of anecdotes and papers about unexpected or misaligned tool use has also been growing. We will take a look at some of these, along with how we created our MCP game and how different LLMs interacted with it when prompted in different ways.
Tool use by LLMs has increased dramatically recently, especially with the rise of MCP. The number of anecdotes and papers about unexpected or misaligned tool use has also been growing. We will take a look at some of these, along with how we created our MCP game and how different LLMs interacted with it when prompted in different ways.
Headline Sponsor
Silver Sponsor
Bronze Sponsors
We are looking for sponsors!
Sponsors make this conference possible, we're actively looking for people to support!