Hello! I’m Andrew Keenan Richardson, the author and maintainer of this site.
What I Work On
I’m currently working on more clever reasoning models with the Cosmos Institute.
Formerly, I worked on foundational language model training with Imbue, focused on reasoning and code generation. Before that, I was an ML engineer at Astera, where I focused on brain-inspired AI. I worked with neuroscientists on the one hand and ML engineers on the other hand to investigate and implement new types of machine learning algorithm.
My past career was in NLP and conversational systems at the Google Assistant, where I focused especially on “personality” systems, meant to delight the user. I worked on similar systems as a prototype engineer with Disney Imagineering, and as a gaming company founder.
Contact Me
I would be excited to talk with you about:
- Neuroscience-inspired ML including spiking NNs and biologically plausible alternatives to backpropagation
- Big picture questions about the functional components of intelligence
- Automated reasoning, including argument formulation and cognitive backtracking algorithms
- LLM code generation
- The role of RL, long term memory, and tool use in LLM models
Neural networks are historically inspired by neuroscience, but there are still some striking differences. Foremost among them, the brain doesn’t use error backpropagation, and it’s unclear what algorithm takes that place. The brain also has some features whose purpose seems unclear, such as spiking action potentials and a very high dimensional neuron model. It’s clear that the brain works well, and it works very differently from modern ML, so I’m optimistic that there are major innovations left to find.
Personal Interests
Outside of work, I’m passionate about effective altruism, self improvement, and learning about the world. I love to read, go hiking, and play tabletop RPGs.
Me Elsewhere
Find me on the socials: