Diversity as the bottleneck in Self-Play
Exploring plateaus in prior self-play setups.
Hamish Ivison
Hamish Ivison is a University of Washington PhD student researching post-training, reinforcement learning, and data for language models.
Latest posts
Introduction to Policy Gradient for LMs
A basic introduction to policy gradient for language models.
Results Replicating L1 for Tulu
Results replicating the recent L1 paper.
My 2025 Reading List
Everything I 'consumed' in 2025.
My 2024 Reading List
Everything I watched, read, and played in 2024.
My 2023 Reading List
Everything I watched, read, and played in 2023.
PhD Hunting 🎯
My own experience around applying for and getting into PhD programs.
Does GPT-3 know Ancient Greek?
Poking around with gpt-3 and ancient languages
Blog Redesign
A quick go-over of my recent blog changes.
AI-ce Attorney
I made a fun little animated Ace Attorney AI script generator.