Google has decided to postpone the launch of its ambitious Gemini AI model, which is expected to rival OpenAI’s GPT-4, to next year. The Information reported this, quoting two sources aware of the matter. They said that Google CEO Sundar Pichai made the call to delay the launch events scheduled for next week in California, New York, and Washington.
Google's multimodal AI competitor
The reason for the postponement is that Gemini, which is supposed to be a multimodal AI model that can understand and generate text, images, and other types of data, is performing poorly in languages other than English. Google engineers found that Gemini lags behind GPT-4 in multilingualism and needs more improvement. Google is already testing smaller versions of Gemini, while the large Gemini model is still under development.
This is not the first time that Google has delayed Gemini. Earlier, it was reported that Google would be pushing back the cloud version of Gemini. Now, the AI-powered products that were supposed to get a Gemini boost, such as the Bard chatbot, will have to wait until next year.
Google announced Gemini at its I/O event, where it claimed that Gemini was “seeing impressive multimodal capabilities not seen in prior models.” Google also said that Gemini was designed to be “highly efficient at tool and API integrations” and to “enable future innovations, like memory and planning.” To attract third-party developers, Google planned to offer Gemini in various sizes, including a lightweight “Gecko” version for mobile devices.
The big question is how soon Gemini will be integrated into Google’s services, such as Bard, Search, and Workspace.
Why Gemini is so important to Google?
The launch of Gemini could be one of the most significant product launches in Google’s history. The AI model is meant to show that Google can compete with or surpass OpenAI while creating a new Internet where the flow of information shifts from traditional search and the world wide web to chatbots.
At the same time, Gemini’s success would be a sign to the industry that OpenAI’s GPT-4 is not the ultimate benchmark but that the underlying Transformer technology and scaling principle (more data, more training) still have room for breakthroughs.
Google has an advantage over OpenAI in data and computing, but so far, it has yet to capitalize on it, partly because Microsoft is partnering with OpenAI.
In fact, since March 2023, no company, big tech or innovative startup, closed source or open source, has been able to release a model close to GPT-4. Instead, the market is flooded with language models at the level of GPT-3.5, a standard that now seems easily achievable.
GPT-4 is so much more advanced because of its larger, more complex, and more expensive architecture. GPT-4 uses many interconnected AI models (A mixture of Experts) rather than a single large model. Google Gemini is likely based on the same concept. OpenAI CEO Sam Altman has already hinted at a timeline for the release of GPT-5, which will be far more advanced.
The complex architecture also comes with a high cost for inference, so OpenAI is trying to lower prices with models like GPT-4 Turbo, even if it means sacrificing some quality.
Originally published on Interesting Engineering : Original article