While ChatGPT has garnered significant attention worldwide, DeepMind, the research laboratory owned by Google, claims that its next large language model will not only compete with OpenAI’s creation but potentially surpass it.
According to an article in Wired, DeepMind is employing techniques used in AlphaGo, their AI system that famously defeated a professional human player in the board game Go, to develop a chatbot named Gemini, which aims to rival ChatGPT.
If all goes according to plan, Gemini will possess the capability to plan, solve problems, and analyze text, as revealed by Demis Hassabis, the CEO of DeepMind, in an interview with Wired’s Will Knight. “At a high level, you can think of Gemini as combining some of the strengths of AlphaGo-type systems with the remarkable language capabilities of the large models,” said Hassabis. He further stated, “We also have some new innovations that are going to be pretty interesting.”
Knight speculates that Gemini, briefly showcased at Google’s I/O developer conference in May, will leverage advancements in reinforcement learning to overcome the challenges faced by current language models when performing tasks. Reinforcement learning involves providing rewards for desired behaviours and penalizing undesired ones, with the objective of teaching the AI system which behaviours to exhibit in specific situations. As Knight highlights, reinforcement learning has already contributed to improvements in the language model domain, playing a key role in how systems like ChatGPT generate responses. Given DeepMind’s extensive experience in reinforcement learning, including the success of AlphaGo, the organization is undoubtedly eager to apply its knowledge to the realm of generative AI.
It’s important to note that Gemini isn’t DeepMind’s initial venture into language models. Last year, the company introduced Sparrow, a chatbot touted to be less likely than other language models to provide “unsafe” or “inappropriate” answers to questions. In an interview with Time in January, Hassabis mentioned the possibility of releasing Sparrow for a private beta sometime this year, although the current status of those plans remains uncertain.
Gemini represents DeepMind’s most ambitious undertaking in the field thus far, particularly considering the involvement of influential figures within Google. The Information reported in March that Gemini, initiated due to the shortcomings of Bard, Google’s previous chatbot project, has received direct support from high-ranking individuals at Google, including Jeff Dean, the company’s most senior AI research executive.
The competition for supremacy in the generative AI domain occurs amidst a surge in investor and customer enthusiasm. According to Grand View Research, the market for generative AI, which encompasses text-analysing AI like Gemini, is projected to reach $109.37 billion by 2030, indicating a significant increase of 35.6% from 2030.