Insights
Writing on data visualisation, AI interpretability, and the craft of building interfaces for complex systems. Part technical, part information design, mostly opinionated.
Machine Learning
Visualising Neural Network Stability
From a simple coffee shop prediction model to exploding gradients, saddle points, and loss barriers — a visual guide to reading the loss landscape and diagnosing instability in neural networks.
Ethics
Fairness is Not A Fact
AGI and bigger models won't be what drives humanity forward — agreeing on what fairness means, and being able to measure it, will be. From capuchin monkeys to COMPAS, this post explores why fairness is the harder and more important problem.
Machine Learning
Visualising Potentially Redundant Capacity
In neural networks, redundancy is about dependent directions rather than unused nodes. PCA and scree plots give us a way to visualise how much unique information each layer is actually using.
Performance
Optimising UI for the new INP Web Vital
The INP Web Vital is a new metric that measures the performance of a website. It is a crucial metric for web developers, as it helps them understand how well their website is performing and how to improve it.
Data
Introduction To Tidy Data Principles
An overview of tidy data principles, emphasizing the importance of structuring datasets for clarity, compatibility, and ease of analysis, with practical guidance on applying these principles using various programming tools.