DeepMind, a subsidiary of Google specializing in AI, is currently developing an algorithm dubbed ‘Gemini’ that will combine natural language processing and reinforcement learning. The goal of the endeavor is to develop a chatbot that exceeds the capabilities of the current leader in the market, OpenAI’s ChatGPT, while also operating more responsibly.
Gemini, Driven by DeepMind’s Expertise in AI Research
Reportedly, Gemini is set to introduce fresh innovations that DeepMind’s CEO and co-founder, Demis Hassabis, finds intriguing. In an interview with WIRED, the CEO asserts that the upcoming iteration will capitalize on the impressive capabilities of the large language models (LLMs).
Hassabis further stated that his team would integrate Gemini with the techniques that enabled AlphaGo to defeat a Go champion in 2016 to endow the system with novel abilities such as problem-solving and planning.
The AlphaGo model demonstrated remarkable mastery over the intricacies of the game of Go, which has an astronomical number of potential moves. Incorporating these techniques could potentially enhance Gemini’s adaptability and versatility beyond ChatGPT, which primarily depends on statistical patterns acquired from extensive text data.
At a high level, you can think of Gemini as combining some of the strengths of AlphaGo-type systems with the amazing language capabilities of the large models. We also have some new innovations that are going to be pretty interesting, the CEO said.