Mini Bio: I am an Assistant Professor at the the University of Cambridge where I lead the CAML Lab. I am also part of the team behind filtir, a fact-checking tool that aims to help tackle hallucinations in AI-generated content. Previously, I was fortunate to be a researcher in the Visual Geometry Group and a research fellow at Balliol College, Oxford.

Research interests

Augmented scientists: The central focus of my current work is developing tools to augment human scientists (by applying machine learning to text, vision and code) to increase their productivity. If you are too, get in touch!

Foundation Models: I have an interest in the recent development of large self-supervised neural networks (sometimes referred to as “foundation models”). See here for some of my online lectures on this topic.

Code generation: As part of the augmented scientists project, I aim to develop tools that allow software development in natural language.

AI safety: This includes both existential risk and pressing near term issues such as the potential for content recommendation algorithms to learn to manipulate humans (“unknowable manipulators”) in a manner that is hard to detect.

Why come to research and build machine learning/augmented scientists in Cambridge? The author of this paragraph is not an impartial judge, but: if you’re passionate about researching and building the machines of the future, Cambridge is the place to be! It has great computing resources, beautiful scenery, and most importantly, a wonderful assembly of smart, creative and energetic people. In the winter it can get a bit chilly, so bring a warm coat.

Other interests

Technical communication: I’m interested in the communication of technical ideas. I have a YouTube channel that represents an exploration on this topic.

Deployment: I have an interest in developing applications that see real-world usage through startups, non-profits etc., particularly those enabled by recent research developments in ML.

Experimental projects:

  • conversations with GPT-4 documents a collection of my interactions with GPT-4 spanning topics such as science, creativity and simulation.
  • samuel-api is a text generation api wrapping a biological large language model.

Feedback is welcome!

A masterfully drawn cartoon depiction of Samuel hiking cheerfully in the mountains. In the background, three gulls hover in an ecologically implausible formation above snowcapped mountains. Samuel carries a stick and wears a knapsack on his back of the kind that is only used to illustrate the Knapsack Problem. The image is signed skillfully, but illegibly. It's the kind of signature that conveys a profound passion for the digital form.