🧠Google Gives Gemini ChatGPT-like Memory

Google’s Gemini chatbot now remembers what you tell it

Key Points 

• The memory feature is currently rolling out to $20-per-month Google One subscribers only.

• For now, this memory feature is exclusive to the web version and supports only English.

☕News - Google’s Gemini chatbot now includes a memory feature that remembers details like your preferences, work habits, and personal info to make conversations more tailored. The feature is gradually rolling out to web users subscribed to the $20-per-month Google One AI Premium plan.

Similar to ChatGPT’s memory, Gemini can use stored details to add context to chats. For instance, if you tell it about your favorite foods, it might suggest restaurants based on your preferences. Other examples provided by Google include reminders to avoid jargon, code in a specific language, or factor in costs for travel planning.

For now, this memory feature is exclusive to the web version and supports only English. Mobile users on iOS and Android will have to wait.

🔏What about privacy concerns? A Google spokesperson clarified that Gemini memories aren’t shared or used to train the model. However, memory features like this aren’t without risks. Earlier this year, a security researcher showed how hackers could implant “false” memories in ChatGPT to harvest user data.

On their part, users can manually delete saved memories or disable the feature entirely if they choose.

Niantic to use data from Pokémon Go players to develop advanced AI navigation system

Key Points 

• Niantic is using data from millions of Pokémon Go players to build an AI navigation system with high precision.

• The AI model will help technology understand and interact with real-world spaces, benefiting AR, robotics, and autonomous systems.

• Unlike Google Maps, Niantic’s system focuses on pedestrian-scale accuracy, providing more detailed data from areas vehicles can't access.

🧭News - Niantic, the company behind popular games like Pokémon Go and Ingress, has announced a new project to create an AI model that can help navigate the real world. Using data from millions of players, they plan to build a system that can map out locations with incredible accuracy.

🕵🏻‍♂️What exactly are they building? Niantic is working on something called a “Large Geospatial Model” (LGM). It's similar to how language models, like GPT, learn from large amounts of text. But instead of learning from text, this model will learn from images and data about real-world places.

The goal is to help computers understand and interact with the world around us, which could be important for things like AR glasses, robots, and self-driving systems. Niantic believes this kind of technology will be key in the future.

🤔What advantage does their AI offer over something like Google Maps? Niantic’s AI has a big advantage when it comes to accuracy because it builds upon its Lightship Visual Positioning System (VPS), which allows players to pin virtual items to physical locations with centimeter-level precision.

So, while Google Maps is great for vehicles, Niantic’s system focuses on areas that only pedestrians can access, providing much more detailed data. As a result, the AI gains a unique perspective of the world, which could be helpful for new technologies like robots or future AR products.