In December 2023, Google launched Gemini, its next-generation multimodal AI model family. Gemini Ultra, the highest-capability variant, was claimed to outperform GPT-4 on a range of benchmarks. Gemini Pro was already available in Bard. Gemini Nano was designed to run on devices.
The benchmarks claims were partially substantiated and partially contested. Gemini Ultra did outperform GPT-4 on some published benchmarks. The methodology used in some of the comparisons was criticised for potentially inflating Gemini’s relative performance. The launch demonstration video was edited in ways that suggested capabilities that did not match the actual real-time behaviour of the model. The reaction was a combination of genuine recognition that Google had produced a competitive frontier model and skepticism about whether the product was as ready as the marketing suggested.
What the launch did establish was that Google could ship a serious frontier model. The internal struggles through 2023, the Bard demo error in February, and the public sense that Google was lagging despite its research history, had created a question about whether Google would catch up at the product level. By December, the answer was clearer. Google was in the frontier model competition and intended to stay there.
The launch closed a year that had been the most consequential in AI history to that point. ChatGPT had crossed a hundred million users. GPT-4 had reset what state-of-the-art looked like. Open source models had become serious. The OpenAI governance crisis had played out publicly. Layoffs had reshaped the technology industry. AI had moved from a research subject to a primary topic of conversation in business strategy, education policy, and public regulation.
What was harder to predict at the end of 2023 was how durable any of the year’s shifts would prove. The pace of capability improvement might continue. It might level off. The cost of running AI applications might continue to drop. It might stabilise as the underlying compute economics caught up. The competitive landscape had been shaken hard enough that the structure that would settle out by the end of 2024 was not obvious.
What was clear was that the AI category had become large enough, and significant enough, that the answers to those questions would be the most important business and technology story of the year that was about to begin.