CursorTab policy gradient optimization

Cursor's new version of the Tab model is now available: online reinforcement learning makes suggestions less accurate and more acceptable

Cursor announced that the new Tab model is the default model: in actual use, the number of tab suggestions is reduced by about 20%, and the probabilit...