It’s possible that ChatGPT got everyone’s attention. However, the Google-owned research facility DeepMind asserts that its upcoming broad language model would be on par with or perhaps superior to OpenAI’s.
A Wired article claims that DeepMind is leveraging AlphaGo’s AI strategies to create Gemini, a chatbot that competes with ChatGPT. AlphaGo was the first AI system to defeat a skilled human player at the board game Go.
According to DeepMind CEO Demis Hassabis, Gemini will be able to plan, solve issues, and interpret language if all goes as planned.
Gemini, according to Hassabis, “at a high level, combines some of the characteristics of AlphaGo-type systems with the incredible language capabilities of the huge models.” We also have some fresh developments that will be rather fascinating.
Knight hypothesises that Gemini, which was briefly hinted at at Google’s I/O developer conference in May, will use advancements in reinforcement learning to complete tasks that current language models find challenging. In order to “train” an AI system which behaviours to display in a specific setting, reinforcement learning entails “rewarding” an AI system for specific behaviours and/or “punishing” undesirable ones.
As Knight points out, improvements in the field of language models have already been made as a result of reinforcement learning, which is crucial to how systems like ChatGPT react to commands. DeepMind is undoubtedly eager to apply its reinforcement learning expertise to the field of generative AI given its amount of knowledge in the field (AlphaGo being one example).
It’s important to note that DeepMind has experimented with language models prior to Gemini. The business unveiled Sparrow last year, a chatbot that the lab claimed was less likely than other language models to respond to queries in a “unsafe” or “inappropriate” manner. It’s unclear if the intentions to release Sparrow for a private beta at some point this year—Hassabis told Time in January—are still on track.
Nevertheless, if early reports are to be accepted, DeepMind’s most ambitious space project to date is Gemini. Gemini, which was inspired by Bard, Google’s chatbot project, failing to stay up with ChatGPT, has direct involvement from Google higher-ups, including Jeff Dean, the company’s most senior AI research officer, according to a March story in The Information.
The competition for supremacy in the generative AI market is fueled by extreme customer and investor enthusiasm. Grand View Research estimates that the market for generative AI, which includes text-analyzing AI like Gemini, could expand by 35.6% from 2030 to $109.37 billion by 2030.