← Videos
Why AI Forgets - Context

AI from the Ground Up8 Nov 2026 · 5:01

Why AI Forgets - Context

You already told the assistant the room name. Why does it get the answer wrong a few messages later?

From the episode description

You already told the assistant the room name. Why does it get the answer wrong a few messages later? “Forgetting” can describe several different failures. A detail may be absent from the current input, lost in a summary, or present but used incorrectly. This episode separates those possibilities through a small, visible example. In this episode: • What tokens represent, and why words are not an exact token count. • What the context window limits. • How an application can select or summarize conversation history. • Why saved memory still needs to reach the current request. • Why a larger context window does not guarantee perfect recall. We compare two summaries of the same fictional Workshop Notes and follow the room name through each. The token splits and diagrams are teaching illustrations, not measurements of a particular model’s capacity or memory performance. The practical question is simple: what information was available for this answer? Use that question to inspect a missed detail, check the source, and decide what the assistant needs supplied again. ResearchNode: AI from the ground up • Episode 02 of 10. Previous: What Actually Happens When You Talk to AI? Next: How to Give AI Better Instructions Presented by Michelle, an AI-generated presenter.

Descriptions and chapter markers are supplied by the channel. They are not a transcript or an independent check of every claim in the video.