Introducing Gemini 1.5
After the launch of Gemini 1.0, Google recently announced its successor Gemini 1.5, a large language model that a user can try now by signing up for a Gemini Advanced subscription. The company is making it available to developers and enterprise users ahead of a full consumer rollout. What makes this AI model apart is that it has a context window of 1 million tokens, the highest of all.
Google makes it clear that this new version of Google’s model Gemini 1.5 is as good as their high-end model, Gemini Ultra which was recently launched. It is better than the previous version, Gemini 1.0 Pro in 87% of benchmark tests. This next-generation Gemini 1.5 model, will use a new Mixture-of-experts (MoE) approach to improve efficiency, which means that instead of processing the entire model every time, a user now can send a query, only part of it runs, making it faster for users to get answers and more efficient for Google to operate.
This model has a big upgrade that has made everyone excited including the CEO Sunder Pichai. This model will handle multiple questions and look at a lot of more information all at once. This new model has a context window of up to 1 million tokens as compared to just 128,000 for OpenAI’s GPT-4 and 32,000 for the previous Gemini Pro, making it far ahead of its competitors as well as predecessors. According to Sundar Pichai: “It’s about 10 or 11 hours of video, tens of thousands of lines of code”. So, a user can ask the AI bot about all of that content at the same time.
Pichai also said that now Google’s researchers are testing a 10 million token context window that is simply like the whole series of Games of Thrones all at once.
New possibilities of Google Gemini 1.5
The larger context window allows users to directly upload large PDFs, code repositories, and even lengthy videos as a prompt in Google AI Studio. Gemini 1.5 pro will then understand different types of information and provide the answers.
Here are some of the possibilities of this New Google Gemini 1.5
Users can Upload Multiple Files and Ask Questions
It can work in the best possible way that a developer can upload multiple files, like PDFs, and ask certain questions in Google AI Studio. This larger context window allows the model to understand information and give an output that is more relevant, useful, and consistent. Thus a user with this 1 million token context window, will be able to load over 700,000 words of text at one time.
Run an Entire Code Via Gemini 1.5
With this model, a user can enable a deep analysis of an entire codebase. This model will grasp more complex relationships, patterns, and understanding of code. A developer will be able to upload a new codebase directly via Google Drive or from his computer and use this model to onboard quickly for the understanding of code.
Add a full-length video on Gemini 1.5
This new language model can work on up to 1 hour of video. Google Ai Studio will break this 1-hour long video into thousands of different frames (without audio), and then a user will perform highly refined reasoning and problem-solving tasks because the Gemini model comes with multimodal dimensions.
Some other specifications of the Gemini 1.5 Model
In addition to bringing the latest model innovations, Google is also making it easier for users to build with Gemini:
Gemini 1.5 Offers an Easy Tuning
A user can instruct this Gemini 1.5 with a set of examples and can customize the Gemini 1.5 for their specific needs in minutes from inside Google AI Studio. This feature will be rolling out soon.
New Developer Surfaces with Gemini 1.5
A user can integrate the Gemini API to build new AI-powered features with the new Firebase Extensions, with the development workspace in Project IDX, or with Google’s newly released AI Dart SDK.
Google is updating the 1.0 Pro model pricing, offering a good balance of cost and performance for many AI tasks. The current stable version is priced 50% less for text- inputs and 25% for outputs than previously announced. Also, upcoming payment plans for using AI Studio on a pay-as-you-go basis will be available soon.
So, since December, the developers have been building multiple projects with Gemini models, and now users can turn their cutting-edge research into early developer products with this Google AI Studio Gemini 1.5.