With the popularity of Chat-GPT and Copilots comes the inevitable speculation of where we are on the AI hype cycle. Are we dealing with "glorified statistics", or is there more to this? In this talk we'll look at the wiring under the board, and what really lies behind Large Language Models like GPT-4. After taking a broad look at Artificial Intelligence and Machine Learning, we'll look at principles like transfer learning, and transformer architectures. As well as the theory, we'll see how it's been applied to solving real problems. And we'll also try to get a feel for where the current AI models sit in relation to human intelligence. Finally, we'll look at the broader AI hype cycle, consider the ethical and existential risks in this rapidly moving field, and see how close we are to artificial general intelligence.

Large language models have been a game changer, but they have their limitations. They only know what was generally available up to the point they were trained. They don't have access to recent events, or your own private data. What's worse, if you ask them about something they don't know about, they start making stuff up! That's why we need to build more sophisticated solutions that combine different models, add some prompt engineering and RAG (Retrieval Augmented Generation), as well as some custom code or plugins, and then a few more layers of safety and responsible AI checks. When you want to combine multiple skills and perform tasks to achieve an objective, you can use Microsoft Agent Framework to handle all the low-level details of handling LLMs. Microsoft Agent Framework is an SDK, available in C# and Python. In this talk, we'll focus on the C# version and build enterprise solutions with .NET. You'll see how we can use it to build an "agent", that goes beyond text responses and that actually does something.