LLMs Don’t Have Memory — So How Do They Remember?

This video explains how memory works in GenAI systems and why it is essential for building chatbots and AI agents. It starts by showing why LLMs are stateless by default, then builds the idea of memory from first principles—covering short-term memory using conversation history and its limitations. The video then introduces long-term memory, explains why it’s needed for personalization and continuity, and breaks down its types (episodic, semantic, and procedural) and how they work at a high level. By the end, you’ll understand how modern GenAI applications add memory around LLMs to make them useful in real-world systems. 📱 Grow with us: CampusX' LinkedIn:   / campusx-official   CampusX on Instagram for daily tips:   / campusx.official   My LinkedIn:   / nitish-singh-03412789   Discord:   / discord   E-mail us at [email protected] ⌚Chapters⌚ 00:00:00 - Introduction: Why Memory Matters in GenAI 00:02:01 - How LLMs Work at Inference (Stateless by Design) 00:09:44 - The Core Problem: LLMs Have No Memory 00:12:28 - Building Memory Around LLMs (First-Principles Approach) 00:13:05 - Context Window and In-Context Learning 00:18:51 - Short-Term Memory Using Conversation History 00:23:43 - How Chatbots Implement Short-Term Memory 00:26:33 - Limitations of Short-Term Memory 00:40:06 - Why We Need Long-Term Memory 00:43:32 - Types of Long-Term Memory (Episodic, Semantic, Procedural) 00:47:20 - How Long-Term Memory Works (High-Level Architecture) 00:53:36 - Challenges and Tools for Memory Systems 00:56:40 - Future of Memory in LLMs 00:57:39 - Closing Thoughts