Gemini: Google's AI Heavyweight Bombshell Claims to Have Completely Surpassed GPT-4's Multimodal Big Model

Beijing, Dec. 7 a.m.googleBreaking Dawn has finally released the long-awaitedAIlarge modelGemini. This mysterious new model is highly anticipated by Google and is seen as a response to theOpenAI(used form a nominal expression)GPT-4The "killer app" of Gemini. Living up to the expectations, Gemini has brought amazing results on its debut: it surpassed human experts for the first time in the MMLU (Massive Multitasking Language Understanding) evaluation, and achieved 30 best-of-breed (SOTA) results in 32 multimodal benchmarks, outperforming the GPT-4 in almost all aspects.

In fact, Gemini's release process has been a bit of a mixed bag. As early as August this year, there were rumors that the Gemini would be released in the fall, however, by November, the release was pushed back to the first quarter of next year. It seems that Google is quite cautious about the development and testing of the Gemini, after all, this is a heavyweight model against the GPT-4.

In this release, Google has introduced three versions for Gemini 1.0: Gemini Ultra for highly complex tasks, Gemini Pro, the best model for a wide range of tasks, and Gemini Nano for end-side devices.Of these, Gemini Pro and Gemini Nano are already integrated on Google's chatbot Bard and smartphone Pixel 8 Pro, respectively. The most powerful Gemini Ultra, on the other hand, is scheduled for release next year.

It is worth mentioning that Gemini is a multimodal macromodel built from scratch. In contrast to the unimodal nature of GPT-4, Gemini recognizes and understands five types of information at the same time: text, images, audio, video, and code. This means that users can naturally input information such as text, images, and short videos, and the model will output them in the same way. This multimodal nature makes Gemini more comfortable in handling complex tasks.

Currently, only the Gemini Pro and Nano are open for users to experience, but according to information released by Google, the Gemini Ultra has already achieved very impressive results on measured benchmarks. In particular, Gemini Ultra's performance is impressive in terms of code generation capabilities. This is undoubtedly a heavy blow to OpenAI at a time when it is mired in internal turmoil.

As technology continues to evolve, the competition in the AI field is getting fiercer.Microsoft corporationHaving just announced a major upgrade of its AI assistant Copilot, which will be connected to OpenAI's latest model GPT-4 Turbo, Google has suddenly revealed the bottom card of Gemini to compete with it with the power of the multimodal large model. With the upcoming release and practical application of Gemini Ultra, we have reason to expect that this multimodal large model will shine in more fields. Let's wait and see if Gemini can continue to set new records and lead the new trend of AI technology.