More thinking makes reasoning models cancel the not all interpretation of some
A test method that uses the first interpretation gives back the not all interpretation at inference time.
Researchers keep the weights of a model the same and change only how long the model thinks. More thinking cancels the usual interpretation of some (not all) and or (not both). This is correct for most open reasoning models that they test, and for GPT-5. Inferences from context, such as indirect answers, are more correct with more thinking. The record of reasoning shows that models first make the interpretation and then cancel it.
Sources
Posted
