Languages and Machine Learning with Richard Stebbing

Vietnamese, GPTs, and 3d dolphins (oh my!)

Today I’m joined by Richard Stebbing who earned his PhD in engineering science with a focus on computer vision from Oxford. We originally met years ago at a private salon of roboticists and ML/AI practitioners who gather in San Francisco to learn about frontier technologies from academic/industry experts.

This episode is all about language and technology — specifically Richard's study of Vietnamese, his interest in large language models, his prior projects in recovering 2D and 3D structures from images and videos, his most recent work on transformer-based models at Impira, and their recent acquisition by Figma and the new areas he's excited to contribute to.

It was a joy to explore these topics with Richard since he is the rare combination of an expert who approaches the world with a beginner's curiosity. I hope you enjoy listening to our conversation.

0:00 - intro

1:07 - learning Vietnamese language

11:11 - learning new programming languages

14:31 - tokenizers

24:19 - hands on with AI programming

28:53 - headwinds and tailwinds in machine learning/AI

45:56 - rooting large language models in factual truth

1:02:53 - Richard's work on 2d/3d computer vision

1:12:27 - Microsoft Research - Trueskill for player matching on Xbox Live

1:24:43 - open source software and the closed/control points in generative AI/ML

1:33:24 - ChatGPT - reopened the conversation of search

1:52:12 - large language models in verticals

1:55:45 - Richard's future work/interests

1:57:42 - Impira - geometric language models

2:11:22 - Figma acquisition of Impira and future of Richard's work at Figma