Agam Bhatia(@chewing_a_gum):We know the J-space can surface what a model is thinking. But can it tell us how a model computes its answer? In a new blogpost, we apply J-Lens to find “meta-tokens” that can directly tell us the algorithm Qwen-3.6-27B uses to complete a task. 🧵

Agam Bhatia

@chewing_a_gum

professional scriptrunner, CS + Math @Stanford

参加 October 2023

1.5K フォロー中 558 ファン

Agam Bhatia@chewing_a_gum

2026.07.21 18:03

We know the J-space can surface what a model is thinking. But can it tell us how a model computes its answer? In a new blogpost, we apply J-Lens to find “meta-tokens” that can directly tell us the algorithm Qwen-3.6-27B uses to complete a task. 🧵