Google has been on a relentless quest to innovate within the realm of artificial intelligence, unveiling a plethora of products that have captivated tech enthusiasts and casual users alike. Among its latest offerings is Gemini, a virtual assistant that integrates seamlessly across various Google devices. However, a lesser-known gem lurking in the shadows of GitHub is the AI Edge Gallery, a unique application that allows users to download and run large language models (LLMs) directly on their Android devices.
What is Google AI Edge Gallery?
AI Edge Gallery is designed for those who seek the power of AI without the need for a constant internet connection. By enabling users to download LLMs, the app allows for offline processing of queries, making it an attractive option for users in remote locations or those with limited connectivity. Currently, the app features four distinct LLMs, categorized into three primary functions: Ask Image, Prompt Lab, and AI Chat.
“AI Edge Gallery allows users to download LLMs that can be used for offline prompt processing right on your phone.”
These categories provide a clear understanding of the app’s capabilities, enabling users to engage in image analysis, chat interactions, and prompt-based tasks with ease.
Installation and setup are a pain, but the app is slick and smooth
The absence of AI Edge Gallery from the Play Store is not without reason; the installation process can be cumbersome. After downloading the app from GitHub, users must create a Hugging Face account and navigate through several user agreements before they can begin downloading the desired models. This setup can be tedious, often leading to frustration.
The setup process from downloading the app to using the model of your choice is padded by several user acknowledgements.
Despite the initial hurdles, the app itself boasts a sleek interface and smooth functionality once operational. I opted for the largest model available, Gemma, which promises comprehensive capabilities across all three categories.
An offline travel planner, science teacher, and sous chef
To evaluate the model’s performance, I posed various prompts typically directed at online services like ChatGPT. My first inquiry was about planning a trip to Spain, where I sought packing advice and sightseeing recommendations. Although the response took a few minutes, I was impressed by the depth of information provided, including specific items to pack and notable attractions to visit.
“Processing times are long, but considering the LLM is running entirely offline on my Pixel 8, it’s admirable.”
Additionally, I tested the model’s ability to explain complex theories, such as Einstein’s theory of relativity, and it delivered a thorough explanation, albeit with a delay. The model also successfully generated recipes and provided insights into various topics, demonstrating its versatility.
How good is the app at creating things?
Within the Prompt Lab section, users can leverage the model to rewrite text, summarize content, and even generate code snippets. For instance, I requested a simple JavaScript code that responds to a specific input, and the model delivered promptly. It also efficiently summarized lengthy texts, making it a valuable tool for processing documents without sharing sensitive information.
However, the tone rewriting feature felt somewhat extraneous, as I found it less applicable to everyday scenarios. Nevertheless, the flexibility of choosing different models based on user needs remains a strong point of the app.
Identifying tomatoes but misplacing monuments
AI Edge Gallery also excels in image recognition. I tested this feature by uploading a photo of grape tomatoes and received a detailed response on how to cultivate them. However, when I inquired about a local landmark, the model misidentified the building, showcasing its limitations in recognizing less common subjects.
“Clearly, Gemma struggles with recognizing buildings but has little issue with tomatoes.”
This inconsistency highlights the model’s strengths and weaknesses, suggesting that while it can be useful in certain contexts, it may not always provide accurate information for every query.
I’ve activated your flashlight (just kidding!)
In terms of functionality, AI Edge Gallery does not possess the same level of control over devices as Gemini. While it can recognize commands, it cannot execute actions like turning on the flashlight or accessing other device features. This limitation is important for users to consider when evaluating the app’s potential as a virtual assistant.
“AI Edge Gallery cannot replace Gemini, at least not as a virtual assistant.”
Despite these limitations, the app offers a glimpse into the future of offline AI processing, providing users with a unique experience that prioritizes privacy and security.
There’s a reason Gemini is Google’s consumer-facing AI product
In conclusion, AI Edge Gallery presents an intriguing option for those interested in offline AI capabilities. While it may not fully replace existing virtual assistants, it offers a valuable alternative for users seeking privacy and independence from constant internet reliance. The app’s potential is evident, and as technology continues to evolve, it may pave the way for more sophisticated offline AI tools in the future.
