Claude's memory beats Gemini's in one crucial way: you can easily control what it remembers - XDA

In my quest to find the one AI assistant that could do it all, I discovered that different models were better at different tasks. If I wanted to code something or break down a complicated task into something easier to understand, I found that Claude was the best at that. However, if I wanted to discuss topics or research something more general, I liked how Gemini handled things.

One thing I didn't like about Gemini was that it didn't remember things between sessions. I noticed Gemini's replies slowly began to unravel the longer a session went; I preferred multiple shorter sessions rather than one big 'master session' on a topic. It kept Gemini's responses 'fresh,' but it also meant I had to repeat myself in the new session.

Then, Google threw me a bone. One day, I went over to Gemini, and Google told me that there was a new feature that allowed the AI to pluck data from our past chats and use them to get a better idea about me. And that sounded wonderful, right up until the point where it wasn't.

Google Gemini's current memory feature is very basic

I kind of hoped for more

What I really like about using Claude is its memory feature. This is a big block of text where Claude details everything it knows about you, including things you might not have known about yourself. It's entirely generated from your interactions with Claude, and it generates a general profile about you based on the information you give it. You can even create separate memories for your projects, which is great if you don't want to mix your recreational Claude with your professional one.

Gemini doesn't have this. From what I can see, the AI's memory is entirely derived from your past chats. There is no master file you can see and tweak like you can with Claude. In fact, if you want Gemini to forget something, it seems you have to delete the chat that the information is in, which isn't great.

Google Gemini kept bringing up things that had nothing to do with the session

It was seemingly forcing links for the sake of building context

So here's the thing. I now have Google Gemini pulling from the tons upon tons of chats I've had with it. There was no truncated file I could look at and edit to ensure Gemini had the right idea about me. In fact, I had no idea what Gemini knew about me, because it could pluck information from any chat I've had with it for the past year or so. That's a lot of personal information.

Despite this, I didn't want to judge the memory feature too harshly before I actually used it, so I opened up some new chats and started talking. That's when Gemini started pulling context from everywhere it could. For instance, I would be talking to it about property prices around my area, and it would suddenly slam on the brakes and ask me if this had anything to do with me visiting the science lab in the city four months ago. And it was cool that it remembered that trip, but at the same time, I didn't want to talk about it right there and then.

The problem was that Gemini knew every little nugget of information I had given it over the past year or so. There was no option within the settings to tweak what Gemini knew, nor was there a master summary file that kept things simple. There was just a toggle that either allowed Gemini to see every iota of information it knew about me or blocked everything. No middle ground.

Claude's master memory file solved this problem before Gemini even had it

I much prefer Anthropic's implementation

Compare this to Claude, which, as I mentioned earlier, doesn't deluge itself with every single chat you've ever had with it to get an image of who you are. Instead, it builds a central memory log that it can refer to at any time.

This implementation beats Gemini's own in two parts. First, because the memory file is more of a summary than a granular description, it keeps the AI from getting bogged down in details. It knows general things about you, which it can pull from when the context requires it, but it's not to the point where it's pulling random months-old factoids.

Second, you can edit the memory file. You can't do it directly, but you can use the memory interface to tell Claude to forget or tweak things to your liking. That way, you can have Claude forget specific parts of you without the need to delete any of your chats. It's not nearly as 'all or nothing' as Google's implementation; it's something you can work on alongside Claude to create something that helps the AI help you.

Gemini's memory needs to be a lot less granular

I understand why Google did what it did. Having Gemini pull from any conversation in the past to generate context in the present is a great way to have the AI help you with tasks without you needing to repeat yourself across sessions. But having Gemini's memory as the chats and not relying on a general summary means that it'll pluck out data that is no longer relevant, and there's not much you can do about it.