dailyai.report

Researchers used J-lens on Qwen3.6-27B to identify "meta-tokens" that trigger specific internal computations. Suppressing the "most likely" token forces the model to commit to single answers instead of hedging. Swapping GCD vectors also altered mathematical outputs. This discovery allows practitioners to steer model logic by manipulating specific vectors in the J-space.