ChatGPT is getting a little upgrade that is sure to be as cool as its controversial. With the flip of a toggle the AI will be able to reference past conversations. That’s crucial if you’re working on a big project across multiple chats, or just looking for a more realistic conversation for a Her-like companion.
It’s also going to be a little alarming, because that’s just more data someone can access if they get your log in or otherwise access your account. But its not unusual either. Both Google and Anthropic already provide similar features for paying customers. In this case OpenAI is just catching up with competitors.
Until today if you were having a conversation with a GPT model what you said stayed in that chat. If you told it you love the color yellow it couldn’t remember that when you opened a new chat and asked it about your favorite color. The only way around that was to toggle on a “Reference saved memories” button and then tell ChatGPT not to forget you love the color yellow.
Starting today, if you’re a ChatGPT Plus or Pro user, you can click a whole new toggle in the same preferences window to “Reference chat history.” Then you can just talk to ChatGPT like usual and it will be able to remember and reference those conversations in future conversations.
I’ve been using this ability in Anthropic’s Claude and Google’s Gemini for a while and I’ve always been throughly pleased with how magical it feels to just have the AI remember all the little things about me–particularly as I was blessed with the family curse of a goldfish memory. When I had a whole conversation with Claude about some new professional plans and then opened a new chat to start brainstorming it immediately knew why I was brainstorming and was quick to help.
But I’d love to see finer control over how these AI access conversations. Right now ChatGPT offers temporary windows that don’t save any of your conversations–essentially the AI version of an incognito browser window. But the button for it is off to the side and easy to miss.
There’s also the constant inescapable feeling you’re giving the AI a lot of details about yourself and your life that you probably shouldn’t. I’ve had those feelings before. Like when I first used Gmail in the 2000s to email saucy fanfiction, and when I uploaded every single one of the photos on my phone to the cloud. The feeling never goes away, but as I’ve used AI more I’ve found the benefit to me personally outweighs the enormous privacy risks. Will that potentially bite me in the ass later? Unclear. When I asked an instance of ChatGPT 4o it said “Honestly, that totally depends on what you’re hoping to get out of this.” Nice to see it and I are on the same page.
This new feature will be available to ChatGPT Plus and Pro users starting today except in the UK, EU, Iceland, Liechtenstein, Norway, and Switzerland. (Not surprising given how much more the EU cares about data retention policies.) The new and improved memory will roll out to Enterprise, Edu, and Team users at a later date. To see if you’ve recieved it keep your eyes peeled for a popup in ChatGPT titled “Introducing new, improved memory” or check in the Settings under Personalization for the new toggle.
