Google’s all in on itsGemini AIbranding, giving its chatbot and Android assistant the name last week, while at the same time rolling out its Gemini 1.0 Ultra model to power it all. While Ultra is currently reserved for developers and Google Cloud customers, regular users can already access Gemini Pro — and today, the company announced that model is being upgraded to Gemini 1.5 Pro, with capabilities that are already meeting or exceeding Gemini 1.0 Ultra’s.
Google Gemini is promising, but it can’t replace Assistant just yet
I tried the ‘experimental AI assistant,’ and my impressions are mixed
In its announcement, Google claims Gemini 1.5 Pro offers “dramatically enhanced performance,” already competing with the higher-tiered Gemini Ultra model in terms of quality, while using less computing power in the process. This is largely made possible with a new Mixture-of-Experts architecture, which allows Gemini 1.5 to only run relevant parts of its model to answer your queries instead of the entire thing each time (viaThe Verge).
But most of the focus of today’s announcement is on Gemini 1.5 Pro’s ability to process huge amounts of information. Google offered the example of feeding Gemini 1.5 Pro a 402-page transcript from the Apollo 11 moon mission, which then allowed it to quickly understand the full story and even identify fun details throughout. You can see this in action in the demo video below.
Jeff Dean, Chief Scientist at Google DeepMind and Google Research, highlighted some of the biggest changesin a Twitter thread. At a maximum, Gemini 1.5 Pro supports 1 million tokens of multimodal input, which is to say you can feed it a lot of information at one time. This means you can get Gemini to analyze and report on things like entire books, document collections, and even full-length movies, opening up a new world of possibilities. While the 1-million token limit will be reserved for early testers at first, a 128,000 token context window will soon be available in Gemini 1.5 Pro’s wider release.
If things seem like they’re moving at a breakneck pace, that’s because they are — Google onlydebuted Gemini a little over two months ago, and it’s already moving into a position where it seems to be comfortable going head to head with the likes of ChatGPT. But the pace of software evolution is expected to continue its acceleration as we get deeper into the AI era, opening up new possibilities in virtually all aspects of life.
