TheChek News Flow

Google has quietly delayed the public debut of Gemini, a conversational artificial intelligence that aims to compete with OpenAI, to January, two people with knowledge of the decision said.

Google CEO Sundar Pichai recently decided to scrap a series of Gemini events, originally scheduled for next week in California, New York and Washington, after the company found the AI didn’t reliably handle some non-English queries, one of these people said. The planned events, which hadn’t been publicized, would have marked Google’s most important product launch of the year, after it strained its computing resources and merged large teams in an urgent pursuit of OpenAI.

• Gemini postponement underscores the challenge of catching OpenAI

• Gemini’s performance is comparable to that of GPT-4 in some ways

• But the model underperformed on some non-English language tasks

The delay underscores how difficult it’s been for Google—which has 200,000 employees and was long considered an AI leader—to catch up to an 800-person startup that has developed one of the fastest-growing software businesses in history. OpenAI also has buoyed Microsoft, a Google Cloud rival that has gotten traction selling OpenAI technology to cloud customers. The Information reported last month that Google told some cloud customers and business partners not to expect to receive access to the primary Gemini model until next year, but the scrapped events also imply that other products such as search (which now has generative AI results), Bard, Google Assistant and Google Docs may not get a boost from the new technology until 2024.

It’s rare for Google to launch a major product between Thanksgiving and the end of the year, but Google intended to make an exception for Gemini because it’s arguably the company’s most important initiative in a decade. The Gemini event in Washington was intended to showcase the technology to policymakers and politicians, which have increasingly discussed potential regulations involving AI, one of the people said. Pichai and other executives have been concerned that OpenAI’s ChatGPT has become a household name and that Microsoft’s Copilot features for productivity software, including for software developers, are turning into a significant business too, people with knowledge of the situation have said.

Companies use the technology, known as large language models, for tasks like automating software coding, summarizing long reports, generating marketing campaigns and building specialized apps that make use of its predictive capabilities.

Last month, Pichai, who in May confirmed The Information’s earlier reports about Gemini, said at a public event that the company is “focused on getting Gemini 1.0 out as soon as possible, make sure it’s competitive, state of the art, and we’ll build from there on.” A Google spokesperson did not immediately have comment for this article.

Google has had all year to develop Gemini, its answer to OpenAI’s most advanced LLM, GPT-4, which debuted in March and is the backbone of the paid version of ChatGPT. Google’s ChatGPT rival, Bard, which uses a more primitive LLM than Gemini, has failed to take off with consumers. That’s bad news for Google because ChatGPT’s millions of customers generate valuable data that helps OpenAI track and improve the quality of its products, and Google is missing out on that opportunity.

Gemini also aims to power new tools for YouTube creators and for advertisers that want to generate text and imagery. Gemini is “mulitmodal,” meaning it can work with images and text alike, producing code for a website just by seeing a sketch of what a user wants the site to look like, for instance, or spitting out a text analysis of visual charts. See the Gemini team here.

Google has developed several versions of Gemini to handle different tasks, depending on their complexity. Outside developers have already tested smaller versions of the model, measured in terms of the number of parameters, or calculations, they do. But the company is still finalizing the primary, biggest version of Gemini, said people who have been involved in the effort. (A big LLM typically is made up of smaller models that work together.)

A key challenge for the Gemini team is making sure the primary model is as good as or better than GPT-4. It has met that standard in some respects, said one of the people familiar with it, but the company is still making improvements because it wants the technology to work well globally, in numerous languages. Model developers frequently test the capabilities of their models while they work on them, but it’s not uncommon for them to miss some of the tech’s limitations during those checks.

OpenAI, meanwhile, has had its own challenges in improving GPT-4 meaningfully, and it quietly canceled a major new model it was developing earlier this year, called Arrakis. But recently OpenAI made a breakthrough, known as Q*, giving the company hope that it can develop a significantly better LLM, The Information reported. The company is still recovering from a near- implosion following the firing and rehiring of CEO Sam Altman.