On the Lunar New Year's Eve, Google came directly with a big vote:Bard will be collectively referred to as Gemini from now on. Through GeminiAdvanced, you can access GeminiUltra, Google's most powerful native multimodal large model!In December last year, Google launched GeminiPro and GeminiNano, which people can use for free through the chatbot Bard, Pixel8Pro and Samsung S24 series mobile phones. Today, Bard has changed its appearance, bringing a new experience of the membership version GeminiAdvanced powered by ultra-large cup Ultra1.0, and also launching mobile applications that support Android and iOS clients.
The new service subscription price is $19.99 per month, which is similar to the price of $20 per month for mainstream generative AI applications such as ChatGPTPlus and PerplexityPro. However, in order to show sincerity, Google will provide a free trial for the first two months.
Google's highest-order multi-modal large model, GeminiUltra opens a new era
According to Google CEO Sundar Pichai, Ultra1.0 is the first model to surpass human experts in MMLU (massive multi-task language understanding), using 57 subject combinations including mathematics, physics, history, law, medicine and ethics to test knowledge and problem-solving abilities.
As a result, GeminiAdvanced will be even more powerful at highly complex tasks such as coding, logical reasoning, following subtle instructions, and collaborating on creative projects. Not only can you have longer, more detailed conversations with your users, but you can also better understand the prompt context.
For example: GeminiAdvanced can become a personal tutor, creating step-by-step instructions, personalized quizzes, or answers tailored to your learning style; solve more complex coding scenarios and help evaluate different programming ideas; become a creative partner for digital creators, generate fresh content, analyze the latest trends, and develop business plans. It’s worth mentioning that all images generated by Imagen2 within GeminiUltra have a digital watermark applied (although you can’t see it).
As new features are added, users will experience greater multimodal capabilities, more interactive programming capabilities, and deeper data analysis. Currently GeminiAdvanced only supports English and can be used in more than 150 countries and regions, and will gradually be expanded to more languages.
Android and IOS mobile versions are available, Gmail, documents, and forms can be used
GeminiAdvanced, as part of the new GoogleOneAI advanced plan, will also provide users with 2TB of storage space. In addition, AI Premium subscribers will soon be able to call Gemini Ultra from Gmail, Docs, Slides, Sheets, and other apps that were previously aggregated as DuetAI.
In order to achieve easy access on mobile phones, Google has launched a new Gemini application this time.
You can snap a photo of a car tire and request a caption, generate a custom image for a dinner invitation, or ask for a complex text message to be written. Google calls it “an important first step in building a true AI assistant—a new class of conversational, multimodal, practical assistant.”
Android phone users can download the Gemini app, or activate Google Assistant the same way you normally would - such as saying "Hey, Google" to wake it up. Gemini can generate a description of the photo you just took and answer questions about the article you are reading. Many of Google Assistant's voice features will also be available through the Gemini App, including setting timers, making phone calls and controlling smart home devices.
Although the iOS app is still on the way, Google said it will be available on the App Store in the next few weeks.
Expand Gemini capabilities to more products
Gemini will also be used across the board in products that individuals and businesses use every day, including Workspace and Google Cloud services.
Workspace:
Pichai said that more than 1 million people are currently using features like "Helpmewrite" to increase productivity and creativity through DuetAI. Starting today, DuetAI will be changed to Gemini for Workspace, and soon, GoogleOneAI premium plan subscribers will be able to use GeminiUltra in the full Google office suite of Gmail, Docs, Sheets, Slides, and Meet.
Google Cloud:
For cloud customers, Gemini will help increase enterprise productivity, assist developers in writing code more efficiently, and protect organizations from cyberattacks.
Developers have been the foundation of every major technological change and play an equally important role in the Gemini ecosystem. Hundreds of thousands of technicians and companies are now using Gemini large models for development. Google will share more details about future benefits for developers and cloud customers next week.
Pichai also revealed that Google is already actively training the next generation Gemini model.
Netizens can’t wait, GeminiUltra is newly launched for testing
When Google released GeminiPro on December 6 last year, it targeted GPT-3.5. Due to the cancellation of the planned offline debut, the Gemini series has been overshadowed by the media. So within a few days, the medium and large cups suddenly appeared together, and triggered heated discussions and online crackdowns on counterfeiting with a shocking "duck" video demo. At that time, Google announced that GeminiUltra had surpassed the industry's most advanced level represented by ChatGPT in 30 of the 32 benchmark tests widely used by LLM.
Now you can finally get your hands on it and find out how powerful the Ultra version is.
Test content generation by writing a LinkedInPost. The conclusion is that GeminiUltra beats GPT-4 and becomes the absolute winner with more title options, faster response times and "no stupid emoji expressions".
Netizen Alphabetting came up with a logical reasoning question: Tabitha likes biscuits but not cakes, likes mutton but not lamb, and likes okra but not pumpkins. It asks, following the same rules, whether Tabitha would prefer cherries or pears.
GeminiUltra suggests: "Tabitha likes foods with two syllables and dislikes foods with one syllable." It lists the number of syllables for each food in the puzzle, and since "cherries" has two syllables, the answer is cherries.
GPT-4 believes that Tabitha's preference may be related to the last letter of the word. The food she likes ends in a consonant, and the food she doesn't like ends in a cause. In this case, both cherries and pears qualify - a bit tricky, but if you have to choose one, let it be cherries.
He said that GeminiUltra successfully solved the logic test that was fragmented by GPT-4.
User Brett Winton tested the Vincentian drawing function of both, and the prompt was "Generate an image of a painter trying to draw a still life on the outside of the rocket to make it humorous, an illustration." On the left is GPT-4, on the right is GeminiUltra.
The imagination of the AI model is a matter of opinion, but the painter in the Gemini image looks more like eating than painting, and the details of his hands are also a bit problematic. Comments all feel that GPT-4 is better.
He then compared the calculation capabilities of GeminiUltra, Claude and GPT-3.5 for 8th grade math problems.
The question is: Garcia is planning a pizza party. She needs to ensure that 30 students each get at least 3 slices, and each pizza has 8 slices. For added variety, Garcia decided to order half cheese pizza and half sausage pizza. However, 5 students are vegetarians and only eat cheese pizza.
Please answer:
1.How many pizzas does Garcia need to order to ensure at least 3 slices for each student?
2. How many pizzas of each type are there?
3. If each pizza costs $12, what is the total order cost?
In previous tests, GeminiPro messed up this question. This time Ultra answered correctly for a total of 12 pizzas and a cost of $144. But the correct answer to question 2 should be 6 of each type of pizza, and Ultra failed.
Brett Winton said that GeminiUltra, like Claude, is not as accurate in mathematical calculations as GPT-3.5.
In terms of coding ability, netizen Mervin Praison successfully created a snake game using Python on GeminiUltra.
For more in-depth use cases, you can take advantage of the two-month free trial and try it out for yourself.
OpenAI opens up a new battlefield for AI agents, and Google is not willing to lag behind
Sissie Hsiao, vice president and general manager of Gemini experience and Google Assistant at Google, said, "For Google, Gemini is more than a model. It is actually a transformation of how we think about the most advanced technology and the entire ecosystem we build on top of it, from products that impact billions of users to the API platform that developers and enterprises use to innovate."
Just yesterday, The Information published a report titled "OpenAI is shifting the focus of AI competition to software that can operate equipment and automate tasks."
The article revealed thatOpenAI is developing an agent software that can effectively take over mobile phones and computers to perform complex tasks for users.You can command ChatGPT to perform data transfers from documents to forms, automate expense reports to be filled and entered into accounting software, or web-based tasks such as creating itineraries or booking flights within a specific budget.
With the launch of more and more new large models, OpenAI is also well aware of the possibility that this year it may no longer have the most powerful LLM on the market. So step up and prepare early to open up new battlefields.
Such requests would trigger agent clicks, cursor movements, text input and other human actions, according to people familiar with the matter. It may turn ChatGPT into what Sam Altman privately calls a "super-intelligent personal work assistant" and will also compete more directly with Microsoft Copilot and Google Gemini for Workspace.
Last year, ChatGPT brought $1.6 billion in revenue to OpenAI, and Microsoft also relied on generative AI to significantly boost its latest quarterly financial results.
The AI business has not yet brought clear cash flow to Google. What kind of market feedback will the launch of paid versions Gemini Advanced and Ultra 1.0 bring? Will those users who have spent money to subscribe to GPT switch to Gemini? What is the new ultimate look of artificial intelligence integrated into the Google ecosystem? These are all exciting. This year is destined to continue to be a tug-of-war for AI with climaxes.