Soon after launch of Gemini, a large language model that Google hopes to bring to the top of the AI industry, the company is already announcing its successor. Today, Google is launching Gemini 1.5 and making it available to developers and enterprise users before a full rollout to consumers. The company is positioning Gemini as a business tool, a personal assistant tool, and a tool for everything in between.
Gemini 1.5 Pro, the general-purpose model in Google’s system, is apparently as good as the Gemini Ultra the company recently released, outperforming Gemini 1.0 Pro by 87% in tests. The model was created using a technique known as «Mixture of Experts» or MoE. This means that it runs only a part of the overall model when you submit a query, rather than processing the query with all its power all the time. This approach should make the model faster to use and more efficient for Google.
Gemini 1.5 has a huge contextual window, which means that the model can handle much larger queries and view much more information at once. This window is a whopping 1 million tokens, compared to 128,000 in GPT-4 and 32,000 in the current Gemini Pro. Google CEO Sundar Pichai explains that: «That’s about 10 or 11 hours of video, tens of thousands of lines of code». The size of the contextual window means that you can ask the AI bot about all this content at once. Pichai also said that Google researchers are testing the context window with 10 million tokens — that’s like all the episodes of «Game of Thrones» at once.
Sundar Pichai also believes that a large contextual window will be very useful for business. «It’s useful in cases where you can add a lot of personal context and information at the time of the request». For example, filmmakers can upload their entire movie and ask Gemini what the reviewers think. Companies will be able to use Gemini to view massive amounts of financial records at once.
For the time being, Gemini 1.5 will be available only to business users and developers through Google’s Vertex AI and AI Studio. Eventually, it will replace Gemini 1.0, and the standard version of Gemini Pro, available to everyone on gemini.google.com and in the company’s apps, will be 1.5 Pro with a context window of 128,000 tokens. You will have to pay extra to get a million. Google is also testing the model’s safety and ethical boundaries, particularly with regard to the new larger contextual window.
Google is now in a race to build the best AI tool as companies around the world try to determine their own AI strategy: whether to sign deals with OpenAI, Google, or someone else. Just this week, OpenAI announced the memory for ChatGPT and is about to introduce its own web search. So far, Gemini seems amazing, especially for those already working in the Google ecosystem, but there is still a lot of work to be done on all sides.
Eventually, all these 1.0 or 1.5 versions, pluses and ultras won’t matter to users. *People will just consume the experience. It’s like using a smartphone without paying attention to the processor in it», — Sundar Pichai. But now, he says, we are still at the stage where everyone knows the chip in their phone because it matters.
Source: The Verge