On December 6, local time, Google released Gemini 1.0, Google’s most powerful general artificial intelligence model to date, once again starting a chase with OpenAI. Gemini 1.0 released this time is divided into three versions: Ultra, Pro and Nano. Ultra is the largest and most capable, capable of handling complex and difficult tasks. Pro is slightly weaker and suitable for expansion of various tasks. Nano focuses more on on-device tasks and can run on mobile phones and other devices.
After the model was released, both Google's official statement and self-media reviews compared Gemini and GPT-4 one by one, with some even bluntly saying that "Gemini beats GPT-4."
Google said GeminiUltra outperformed OpenAI’s GPT-4 in a series of tests conducted against industry standards. Specifically, from natural images, audio and video understanding to mathematical reasoning, GeminiUltra's performance exceeded the current advanced level in 30 of 32 academic benchmarks. In terms of MMLU (large-scale multi-task language understanding), GeminiUltra scored as high as 90.0%, surpassing human experts for the first time.
Can Gemini really crush GPT-4?
After collecting and comparing information, it is not difficult to find that Gemini’s release contains a lot of moisture. Whether GeminiUltra and GPT-4 can really compete, we still need to wait until GeminiUltra is officially launched and applied before we can test it ourselves.
In fact, Google’s ambition to catch up with OpenAI has never stopped.
In February 2023, Google launched Bard, a conversational generation AI chatbot powered by LaMDA. But soon, netizens discovered that there were incorrect answers in the Bard case officially announced by Google. Google's stock price also fell by more than 5% after the market opened on the day the news was fermented, and Microsoft once rose by more than 3%.
I wonder if Gemini this time can really bring about a comeback for Google?