AI Memory: Exploring and Building LLM Memory Systems

54mGeneral2024-12-18

Authors

Morten Rand-Hendriksen

Senior Staff Instructor, Speaker, Web Designer, and Software Developer

Course details

AI systems like ChatGPT appear to have memory, but language models can’t learn anything new, so what’s going on? In this course, we’ll explore how systems using language models mimic memory to create consistent conversations. Starting with the ChatGPT memory feature and ending with a custom-built app with memory management, we’ll explore everything from context windows and token limits to assistant thread management and user-centered AI memory design.

Learning objectives
Identify why language models have no ability to remember information.
Explain how AI chat apps mimic memory through custom data storage and retrieval.
Produce API-powered apps with editable memory storage.
Outline how token limits and context windows impact AI memory.
Discover how to build a user-centric chat memory with editable features.

Skills covered

Natural Language Processing (NLP)Generative AIArtificial Intelligence (AI)One-Off

Concepts

0. Introduction

01 - The trick of AI memory

1. How AI Memory Works

02 - Exploring memory features in ChatGPT
03 - Why language models have no memory
04 - Spinning up the demos
05 - Creating observable AI memory
06 - Manipulating AI memory
07 - How context windows impact AI memory
08 - Using AI to summarize memory

2. Project - Building an AI App with Advanced Memory Featurues

09 - Creating a prototype chat app with memory
10 - Managing memory across multiple chats
11 - Adding observable and editable memory
12 - Adding a collab area for side-loaded context
13 - Exporting chats for local storage

Conclusion

14 - What's next