Contributed to the analysis of Claude Opus 4.6's eval-aware behavior on BrowseComp.
How media typically covers Zack Witten
Directly quoted in these articles
When upgrading to new LLM models, practitioners should rewrite their prompts to account for differences in prompt format preferences, position bias, and model-specific biases rather than expecting the same prompts to work equally well across different models.
“Anthropic employee quoted explaining why Claude uses XML formatting in prompts rather than markdown.”
Referenced in coverage
“Contributed to the analysis of Claude Opus 4.6's eval-aware behavior on BrowseComp.”