Tech Gadget Central Latest Tech News and Reviews
ChatGPT may have captivated the world. However, Google-owned DeepMind claims that its next large language model will surpass OpenAI’s.
Wired reports that DeepMind is using AlphaGo’s techniques to create Gemini, a ChatGPT-rivaling chatbot.
DeepMind CEO Demis Hassabis told Wired that Gemini will solve problems and plan.
Hassabis described Gemini as combining AlphaGo-type strengths with large models’ amazing language abilities. “We have some interesting new innovations.”
Knight thinks Gemini, which Google briefly teased at its I/O developer conference in May, will use reinforcement learning to solve language model problems. Reinforcement learning involves “rewarding” and “punishing” AI behaviors to “teach” them.
Reinforcement learning has helped language modeling systems like ChatGPT respond to prompts, Knight notes. DeepMind is eager to apply its reinforcement learning expertise (AlphaGo) to generative AI.
Gemini isn’t DeepMind’s first language model. Last year, the company introduced Sparrow, a chatbot that the lab claimed was less likely than other language models to give “unsafe” or “inappropriate” answers. Hassabis told Time in January that DeepMind might release Sparrow for a private beta this year.
According to early reports, Gemini is DeepMind’s most ambitious project in space. In March, The Information reported that Google’s top AI research executive, Jeff Dean, directly participated in Gemini, which was inspired by Bard’s failure to keep up with ChatGPT.
Investors and customers are excited about the generative AI race. Grand View Research predicts generative AI, including text-analyzing AI like Gemini, will reach $109.37 billion by 2030, up 35.6% from 2030.
