J Studios/Getty Images
ZDNET’s key takeaways
- Claude can now be prompted to reference past user interactions.
- The feature rolls out today to Max, Team, and Enterprise users.
- It’ll be turned on by default, but you can also switch it off.
Claude just got a major memory upgrade: Anthropic’s flagship generative AI chatbot can now retrieve information from past conversations, the company announced Monday. The new feature is designed to enable a more streamlined, convenient, and personalized user experience.
Also: Claude Sonnet’s memory gets a big boost with 1M tokens of context
If you’re collaborating with the chatbot on a project and need to take a break midway through, for example, you can come back and simply prompt it to pick up where you left off. In a YouTube video posted by Anthropic, a user tells the chatbot that they’ve just returned from vacation and asks for a reminder of what they’d been working on before they left. The chatbot is shown organizing the content of its conversations with the user from that time period by subject, narrowing the list down until it’s able to clearly convey the specific task the user had been engaged in before their vacation, and suggesting a couple of next steps.
Users can now prompt Claude to pull information from all of their previous conversations, or from a specific project.
How much memory is enough – or too much?
Frontier AI companies like Anthropic and OpenAI are racing to build more reliable and sophisticated chatbots — the ability to remember information over time has become an important component in that competition.
Also: Why AI chatbots make bad teachers – and how teachers can exploit that weakness
It’s a delicate dance. Allow a chatbot to remember too much information about a particular user, and you start to veer into privacy issues or risk mental health downsides; a user could, for example, run a higher risk of becoming emotionally dependent upon a chatbot that seems to know and understand them, as opposed to one that just delivers generic and impersonal responses. But constrain their memory too much, and they start to fall behind competing chatbots, which are constantly becoming smarter, more agentic, and more fine-tuned to the needs of individual users.
ChatGPT, for example, can form “memories” of important biographical details about users, like their name or profession, that will help inform its future responses (though this feature can also be turned off). Users can also directly ask the chatbot to remember important personal details. All saved memories can then be viewed and, if necessary, deleted, giving users a measure of control over how granular and comprehensive a user profile the chatbot builds for them.
The basic idea is that, by recalling information from previous chats, ChatGPT will gradually be able to deliver a more personalized experience to each user. “The more you use ChatGPT, the more useful it becomes,” OpenAI writes in the FAQ section of its website. “You’ll start to notice improvements over time as it builds a better understanding of what works best for you.”
Also: How to use ChatGPT freely without giving up your privacy – with one simple trick
Anthropic is taking a slightly different course with its new upgrade for Claude, allowing the chatbot to only retrieve information from past conversations when it’s been explicitly prompted to do so.
How to access the new feature
Accessible via desktop, mobile, and the Claude app, the new memory feature is being rolled out today to paid subscribers of the company’s Max, Team, and Enterprise plans. It’s expected to be made available to more users soon.
Also: I tested this new AI podcast tool to see if it can beat NotebookLM – here’s how it did
Once it’s been made available to your subscription tier, the memory feature will be turned on by default, according to Anthropic’s website. Users also have the option to disable it by going to Settings > Profile > Preferences and toggling “Search and reference chats” to off.
