Running an LLM from scratch - A detailed introduction

In this session we introduce simply, from a technical perspective, what LLMs actually are and how you can actually run one both for yourself and your firm on private infrastructure. The idea is to demystify what is often portrayed as a very complicated subject that often prevents legal practitioners from making informed decisions when picking providers. The aim of this masterclass is simple: 1) Demonstrate how to get a rented GPU in the EU 2) Actually show the contents of a model i.e. fully show what consists of a model 3) Run a small SATA (state of the art) model open source model live (this will either be a Gemma or Qwen model). This will not be an ollama type run but an actual production-ready deployment of a model. 4) Demonstrate that models can be run entirely privately & discuss self learning models. 5) Demonstrate what we mean when we say that these tools are statistical in nature. This involves showing where the probabilities actually feature on the interaction with the model This should be complete within the first 15 minutes. After this, I will let participants actually interact with the model and have a Q&A session. The idea is that you leave the room with a clear understanding of what an LLM actually is. This should remove a lot of the fluff from many of the current presentations & pitches that mention AI as an amorphous almost magical tool. My aim is to have you think of AI as a tool first.