Surpassing GPT-4 is difficult? Google's "most powerful model" release postponed to the first quarter of next year.
Google's large-scale model Gemini has been delayed until the first quarter of next year, and the development difficulty may be greater than expected.
In the arms race of big models among Silicon Valley tech giants, Alphabet-C seems to be falling further behind its competitors OpenAI and Microsoft.
According to media reports on the 16th, Alphabet-C's Gemini big model, originally planned to be launched before November, has been delayed until the first quarter of next year.
It turns out that catching up with OpenAI-4 is not an easy task.
Alphabet-C's cloud revenue growth has slowed down, while its competitor Microsoft is accelerating its growth. Part of Microsoft's success comes from selling OpenAI's OpenAI big model to customers.
Since the launch of OpenAI a year ago, this startup has reached the pinnacle of the tech industry and has made AI chatbots synonymous with artificial intelligence in the minds of global consumers. What embarrasses Alphabet-C the most is that OpenAI has created OpenAI by using technology invented by Alphabet-C and hiring former Alphabet-C employees.
Gemini can be said to be Alphabet-C's battle to regain the title of Silicon Valley's AI leader. It is highly anticipated by the company, and co-founder Sergey Brin has even returned to the office to personally write code.
Alphabet-C hopes that Gemini can drive revenue for its cloud services and provide new tools for creators on YouTube, such as the ability to generate custom backgrounds for videos, and improve the capabilities of Bard and Google Assistant (Alphabet-C's voice assistant similar to Siri for phones and other devices).
During the earnings call on October 24th, Alphabet-C CEO Sundar Pichai did not discuss when Gemini would be launched. He stated that the company is still laying the foundation for "next-generation big models" and will launch these models in 2024.
Alphabet-C has developed multiple versions of Gemini to handle different tasks based on their complexity. External developers have tested smaller versions of the model, measuring them by the number of parameters or computations.
However, insiders have told the media that Alphabet-C is still developing the main and largest version of Gemini. One key challenge the team faces is ensuring that the model is as good as, or even better than, OpenAI's most advanced LLM (OpenAI-4). The insider said it is still unclear whether Alphabet-C has reached this standard.
While Alphabet-C's Gemini is being delayed, OpenAI is making great strides. On November 7th, at OpenAI's first Developer Day, OpenAI-4 Turbo, Custom OpenAI, and the OpenAI Store were launched one after another. After going live, they quickly attracted eager users, leading to a surge in the usage of OpenAI Plus. New member users have overwhelmed the servers, and OpenAI has even had to temporarily halt Plus membership registrations.