Now

A living document of what I'm currently working on, exploring, and thinking about. Inspired by nownownow.com

Last updated: January 6, 2025

Currently Exploring

Mixture of Agents

Testing if routing queries to specialized models outperforms a single large model.

Claude's new computer use

Building automation workflows with Claude's computer use capabilities.

Evals-driven development

Creating comprehensive evaluation suites before building features.

Reading

Scaling Monosemanticity

Anthropic's research on understanding features in Claude

The AI Engineer Newsletter

Weekly updates on the AI engineering landscape

Open Questions

Things I'm actively trying to figure out. If you have insights, I'd love to hear them.

  • ?When does fine-tuning beat in-context learning for domain adaptation?
  • ?What's the optimal agent architecture for long-running tasks?
  • ?How do we measure "reasoning" in LLMs beyond benchmark gaming?