ChatGPT just got a memory upgrade that works more like a human brain — and it could fundamentally change how you interact with AI every single day.
What Is 'Dreaming' and Why Does It Matter for AI Memory?
OpenAI has introduced a new memory system for ChatGPT called "Dreaming" — a process that consolidates and refreshes what the model knows about you between conversations. Think of it like the way your brain processes and organises the day's events during sleep.
Instead of storing raw conversation logs or relying on you to manually save preferences, the system synthesises patterns from past interactions into a cleaner, more useful memory. The result: a ChatGPT that actually gets better at helping you, specifically, over time.
How the New AI Memory System Works in Practice
Previous memory in ChatGPT was fairly blunt — it remembered facts you explicitly told it, like your name or job title. Dreaming goes deeper, picking up on implicit preferences: how you like information structured, your communication style, recurring topics you care about.
Crucially, the system is designed to keep context "fresh and relevant" — meaning stale or outdated preferences get deprioritised rather than cluttering future responses. It's less a filing cabinet, more a living profile that evolves as you do.
This is a meaningful step toward truly personalised AI — the kind that doesn't make you repeat yourself every session. If you've ever wished your AI assistant had a better memory, this is the breakthrough you've been waiting for.
What This Means for Learners
Personalised memory changes the game for anyone using AI as a learning tool. A ChatGPT that remembers your skill level, preferred explanation style, and learning goals can scaffold knowledge far more effectively than a blank-slate model.
To get the most out of systems like this, you need to understand how language models store and process context in the first place. Our Decoding Language Models Tokenization course explains exactly how context windows and memory interact under the hood — essential knowledge for power users.
And if you want to go further and build AI tools that remember users across sessions yourself, Build Your First RAG Pipeline teaches you the retrieval-augmented architecture that makes persistent, personalised AI possible.
