During the Google I/O presentation, the company unveiled an updated flagship model of artificial intelligence, Gemini 1.5 Pro, which can analyze even larger documents, code bases, videos, and audio recordings, writes TechCrunch..
Gemini 1.5 Pro can accept up to 2 million tokens, separated by bits of raw data, such as the fan, tas, and tic syllables in the word fantastic. This is also double the capacity of the previous model. A total of 2 million tokens is equivalent to about 1.4 million words, two hours of video, or 22 hours of audio.
If 1 million tokens is a lot, how about 2 million?
Today we’re expanding the context window for Gemini 1.5 Pro to 2 million tokens and making it available for developers in private preview. It’s the next step towards the ultimate goal of infinite context. #GoogleIO pic.twitter.com/3OW77YH4Ec
— Google (@Google) May 14, 2024
The large number of tokens also improves the overall performance of the artificial intelligence. Unlike smaller models, 1.5 Pro won’t forget the context of previous conversations as quickly, and thus won’t jump from topic to topic. It also has the potential to provide better-generated and more contextualized answers to questions.
Google also claims that the 1.5 Pro model has undergone many improvements over the course of several months due to algorithm enhancements. The company says that the model has become better at creating code, logical reasoning, planning, multi-way conversation, and understanding audio and images.
In addition, Google also introduced Gemini 1.5 Flash, a smaller and more efficient model that was created for narrow and high-frequency AI generative workloads. Despite this, Flash can still analyze text, video, audio, and images just like Pro, but generate only text.
Introducing Gemini 1.5 Flash ⚡
It’s a lighter-weight model, optimized for tasks where low latency and cost matter most. Starting today, developers can use it with up to 1 million tokens in Google AI Studio and Vertex AI. #GoogleIO pic.twitter.com/I1adecF9UT
— Google (@Google) May 14, 2024
In general, Gemini 1.5 Flash is better suited for tasks such as summarizing text, chat messages, analyzing video and image captions, and searching for data in large documents and tables.
“Gemini Pro is for much more general or complex, often multi-step reasoning tasks. [But] as a developer, you really want to use [Flash] if you care a lot about the speed of the model output,” said Josh Woodward, vice president of Google Labs.
Among other things, both Gemini models will also soon receive a context cache, which will allow developers to store large amounts of information in the model’s cache so that it can access this data more quickly later.
Loading comments …