ChatGPT remembers. OpenAI’s chatbot has memory capabilities that allow it to recall information about users for personalized responses. This interesting feature, which is optional, is now also available on another popular platform: Google Gemini. However, memory comes at a cost.
Memory. The similarities to ChatGPT are evident, given that Gemini operates on a similar system. Google’s AI chatbot can remember specific details users have shared, including personal preferences, formats, dates, and styles, and will apply this information to future responses. The goal is for Gemini to increasingly adapt to the user’s individual needs.
How it works. For example, if a user informs Google Gemini that they’re celiac, the chatbot will remember this piece of information. Then, when they later ask for a recipe or dish suggestion, Gemini will avoid suggesting any dishes containing wheat. Meanwhile, if you tell the chatbot your name and mention that you work at a specific company, the next time you request assistance with an email, it’ll introduce you by saying, “I’m [your name], and I work at [your company].”
False memories. Although Gemini will learn as it goes along, users can “force” it to remember specific information by using the Saved Info page. This feature allows you to add details for the chatbot to consider. You can also view, edit, and delete the information stored.
A paid feature, for now. Similar to ChatGPT, accessing Google Gemini’s memory requires a subscription to Google Gemini Advanced, which costs $19.99 per month. This plan includes the memory feature, access to Gemini 1.5 Pro, a context window of up to one million tokens, and 2 TB of storage in Google One.
Persistent memory. Gemini’s new memory feature is an intriguing advancement, but it’s still in its early stages. Currently, AI systems possess a form of short-term memory, which is the type Google has introduced in its chatbot. The next significant milestone for AI development is achieving persistent memory.
Mustafa Suleyman, a co-founder of DeepMind and the CEO of Microsoft AI, recently explained that this represents the second of three major phases in AI evolution. The first phase involves AI systems processing information similarly to humans, specifically, seeing, hearing, and interpreting the world as we do.
The second phase is about enabling AI chatbots to create shared knowledge and understanding. This will allow them to engage in conversations and interactions that are enriched by extensive data, experience, and learned knowledge. In other words, this phase focuses on developing memory.
The third phase involves creating agents, which are AI systems capable of interacting with third parties. This means that the AI model could perform tasks for us, such as purchasing items and booking flights, with minimal input beyond a simple instruction. While phases two and three are currently in development, they still have a significant way to go before becoming fully realized.
Image | Luca Merz
View 0 comments