Naomi Bashkansky's Technical Blog

Where I write about my projects, paper replications, and technical explainers.

Sparse Autoencoders for a More Interpretable RLHF

Extending Anthropic's recent monosemanticity results toward a new, more interpretable way to fine-tune.

1 min read · December 12, 2023

2023
Examining Llama 2's Propensity for Following the System Prompt

Large Language Models seem to 'forget' their system prompt over the course of a long conversation. Can we measure this effect?

1 min read · December 7, 2023

2023