To compete against ChatGPT, Google is finally upgrading its Bard AI with the company’s highly anticipated “Gemini” model, which promises to make the chatbot more capable than ever.
Google is releasing the Gemini model in three versions: The first, Gemini Pro, will arrive through the free Bard service, first for English users. And according to Google, the model has made Bard smart enough to beat OpenAI’s free ChatGPT service in six out of eight benchmarks, including in math and language understanding.
“This is the biggest upgrade to Bard since it launched,” the company says. In the coming months, users can also expect Gemini to power other services, including Google Search, ads, and Chrome.
(Credit: Google)The second version, Gemini Nano, is small enough that it can run locally on a smartphone. The company is rolling out the Gemini Nano model to Pixel 8 Pro device. But it won’t be running a native version of Bard. Instead, Google is using Gemini Nano on the Pixel 8 Pro to power the Summarize feature in the Recorder app and the smart reply function in Google Keyboard.
Meanwhile, the most powerful model, Gemini Ultra, can outperform OpenAI’s latest GPT-4 model, used for the paid ChatGPT Plus service, according to Google’s benchmarks. But the company still needs to test the model to ensure it's trustworthy and safe to use. As a result, Gemini Ultra won’t arrive until early next year through a new tier called “Bard Advanced,” which will give users access to the most cutting-edge AI models. But whether it’ll cost users money remains unclear.
(Credit: Google)In the meantime, Google is demoing Gemini’s capabilities on a website that shows the AI model can understand user inputs in a variety of mediums, including text, images, audio, and computer code. For example, the model can produce a small computer program simulating the movement of a flock of birds simply by showing Gemini a video.
Another demo shows that Gemini can understand math problems on a piece of paper and supply the correct answers. It can also go through hundreds of thousands of scientific papers and extract insights the user is looking for all within a lunch break.
Along with users, Google plans on making Gemini models available to third-party developers building apps or internet services. The access will arrive on Dec. 13 through Google AI Studio and Google Cloud Vertex.