Notebook · 2019–2026
Research logs, artifact notes, and a small archive from the old site. Some entries are polished; others preserve the state of an idea while it was still being tested.
FCE beat a permuted-reward control by 20.8 points across four SmolLM seeds and raised Qwen2.5-1.5B from 54.6% to 66.8% on GSM8K. Majority, harder-domain transfer, and unanimous-wrong-consensus claims remain explicit non-wins.
A displacement-controlled route from row-energy conditioning back to canonical Muon, plus the harder question of extending the idea to coupled attention circuits.
Use the orthogonal SVD basis, select input-specific components with OMP, and compare against trained decomposition methods without hiding the unsupported causal claims.
A small system where models propose Python solutions and a separate deterministic validator searches for counterexamples before accepting them.
An explicitly exploratory model-internal signal tested in the Apart AI Control Hackathon, with its evidence boundary kept visible.
An interactive investigation of attention sinks, over-mixing, and distributed sink banks built as a Marimo notebook.
An evolving list of free and open-source tools for research, writing, and focused work.
Notes from entering a PhD without a fixed research area, then learning how to read, write, collaborate, and choose problems.
An entry point to the theory, implementations, and deployment questions around differential privacy and federated learning.
A practical guide to collaboration, documentation, benchmarking, and reproducible machine-learning experiments.
A short personal essay about the inner voice, attention, and returning to the small things in daily life.