SEEING EYE TO AI- HUMAN ALIGNMENT VIA GAZE-BASEDR ESPONSE REWARDS FOR LARGE LANGUAGE MODELS