Models Know What to Change. But Do They Know What to Leave Alone?

Automated excerpt

My first evaluation gave DeepSeek, Claude, and Gemini perfect reported scores across ten cases each. Testing whether a model changes only the requested setting is an important problem in its own right. The initial long-form evaluation produced perfect reported accuracy across all ten cases for each model.

Selected automatically from source text; not independently written or fact-checked. Read the original for full context.