Gemini Live Introduces New Eye Feature: Our Hands-On Review
Exploring Google’s Gemini Live: A New Way to Interact with AI
Introduction to Gemini Live
Recently, I had the opportunity to test out Google’s Gemini Live feature while giving a tour of my apartment. Using my phone, I engaged the AI by showing it different objects and asking it questions. For instance, when I pointed out a pair of scissors, Gemini Live instantly identified their location on my table, right next to a green package of pistachios.
What Can Gemini Live Do?
Gemini Live can recognize a wide variety of items beyond simple household objects. Google claims that this AI can assist users in various situations, such as:
- Navigating in crowded places like train stations.
- Determining what’s inside a pastry by analyzing its appearance.
- Providing context and history about artwork, such as its origin and limited edition status.
This interface moves beyond traditional AI features; it allows for a two-way conversation. Unlike the older Google Assistant, Gemini Live responds to users in a more natural, conversational style.
Availability of Gemini Live
Currently, Gemini Live is being launched for Google Pixel 9, including the new Pixel 9a, as well as Samsung’s Galaxy S25 phones. Users of these devices can access the feature for free. Users with other Pixel models can try Gemini Live through a Google AI Premium subscription. For detailed tutorials, Google has released a YouTube demonstration that showcases this functionality.
To start using Gemini Live, all you need to do is activate the camera and engage in a conversation with it.
The Development Behind Gemini Live
Gemini Live is a continuation of Google’s Project Astra, which was presented last year as a forward-looking feature in the company’s generative AI capabilities. This lines up with the advancements being made by various tech companies in enhancing AI usability, including Apple’s Visual Intelligence technology.
Real-World Applications: Testing Gemini Live
Despite the early rollout, I had the chance to test Gemini Live on my Pixel 9 Pro XL before its official release. My first challenge involved showing Gemini a specific gaming collectible, a plush rabbit, which it recognized with surprising accuracy. The next test took place in an art gallery where it not only identified a tortoise depicted in a piece of art but also translated nearby kanji characters, which was both impressive and somewhat eerie.
Experimenting in My Apartment
During my apartment tour, I attempted to tax Gemini’s object recognition capabilities with various everyday items. From fruit to cosmetics, the AI often got them right. Encouraged by this success, I decided to test its limits further, especially with my collection of horror-themed memorabilia.
Initially, I encountered some hiccups, particularly when trying to record my sessions. However, when focusing on specific items from my horror collection, I found that Gemini was both surprisingly accurate and at times amusingly off-base.
Challenges with Recognition
Notably, Gemini Live showed inconsistent performance during testing. On occasion, it would nail specific queries about more common or popular items, such as collectibles from the game "Destiny 2." Other times, it would misidentify objects and struggle with providing accurate details. For instance, it performed exceptionally well when identifying a limited-edition gaming figurine from a well-known series, yet misidentified a more obscure collectible.
As the testing sessions extended, I noticed that the AI’s accuracy sometimes declined. The AI appeared to draw conclusions based on previously identified items, which led to further confusion. This feature may allow it to connect objects in unexpected ways, but it can also lead to multiple incorrect guesses in succession.
Key Insights from Interaction
Throughout my interactions, I saw that providing hints could be beneficial, but not always reliable. Sometimes, the prompts I offered helped steer Gemini towards the correct answer, while other times, it would continue to repeat incorrect responses. It was interesting to note that conversations often performed better when initiated from past successful interactions.
In my various attempts to identify items—from beloved video game characters to artwork—Gemini Live exhibited a mix of impressive recognition and head-scratching mistakes. Its dual nature serves as a reminder of the complexity and potential of AI technologies in our everyday lives.
