Build the thing now.
So we can build the future.
The session moves from seeing the inside of a small language model to training, testing, building an agent, and shipping a live website.
What happens in the room
See how a model works
Trace one word through a real language model. See tokens, weights, and the scores behind a prediction.
Train it twice
Choose a dataset, train a model, corrupt the data deliberately, train again, and compare the results without labels.
Make an agent for your own use
Choose what you want it to do during the session, connect a tool, and test how it handles your task.
Ship a personal website
Deploy a website to a live address before the session ends. Bring the result home.
Bad data makes a model unreliable.
The two training runs are an experiment participants conduct themselves. A corrupted dataset may not teach a clean lie; it can also damage what the model appeared to know. That is a harder failure to spot than a simple wrong answer.
A laptop and curiosity.
No prior coding
The room is designed for someone who has never written a line of code.
No installs
The workshop uses tools in your browser and our computing infrastructure.
No graphics card
A personal GPU is not required.
Something to question
Participants choose data and test the model's answers for themselves.
Bring this workshop to your room.
Tell us about your room and we’ll reply with the right format.
Send an enquiry