video1.0<iframe src="https://www.loom.com/embed/b92f46b5d22e44b181b755af1a979092" frameborder="0" width="1662" height="1246" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>12461662Loomhttps://www.loom.com12461662https://cdn.loom.com/sessions/thumbnails/b92f46b5d22e44b181b755af1a979092-4955b2368f61a235.gif59.199This Loom explains how an eye-tracking system helps LLMs tailor responses to what the user has actually read and consumed. The speaker describes building an ontology for Reddit applications so the system can track specific facts a user has read, such as noting that MongoDB was first publicly released in 2009. This consumption history is then prepended to the following prompt so the LLM does not assume the user read all prior context. The goal is to reduce unnecessary output that many users do not read.