← Browse

Session brief

Video Has No Memory. Here's How We Built One. — James Le, TwelveLabs

James Le

20 min

Overview

This talk addresses the challenge of enabling AI models to retain information across extended video content, effectively giving them memory. The presenter discusses the technical hurdles and the architectural solutions developed to overcome these limitations, allowing for more sophisticated video analysis and interaction.

Who should watch

Key takeaways

Notable quotes

*AI models do not inherently remember past information without specific design.*
*Building memory into AI systems is key to unlocking advanced video understanding.*

Watch on YouTube →

Unofficial community note. Prefer the recording for nuance.