DeepMind, the Google-owned research lab, claims its next big language model, called Gemini, will rival or even will outperform OpenAI’s ChatGPT.
No one wants to be left out of AI research once the launch of ChatGPT as the chatbot of fashion, together with the neural network that supports it and is said to be the most powerful on the planet, has shown the potential of AI and has made it become an emerging market for the entire technology sector. The potential potential is vast, but also its risks and problems.
DeepMind, the Artificial Intelligence division of Google, wants to use AlphaGo techniques to create a new chatbot with the potential to “outshine” ChatGPT. If you don’t know him, say that he was the first AI to win the world champion of the board game Go. Development has focused until now on competing in others such as checkers, chess or poker and also on complex video games with the aim of beating human players in the multiplayer mode of Starcraft II, the real-time strategy video game.
DeepMind wants to take advantage of all that experience and the resulting neural network to create Gemini, a chatbot that will have the ability to plan or solve problems, as well as analyze text. “At a high level, you can think of Gemini as a combination of some of the strengths of AlphaGo-like systems with the amazing language capabilities of large models and some innovations that are going to be quite interesting”explains the CEO of DeepMind.
The executive explains that Gemini take advantage of innovations in reinforcement learning to perform tasks that current language models struggle with. Reinforcement learning involves “rewarding” an AI system for certain behaviors and/or punishing unwanted ones, with the goal of “teaching” the system which behaviors to exhibit in a given situation.
Reinforcement learning has already shown its advantages in language models and is key for systems like ChatGPT to respond to prompts. DeepMind has a wealth of experience in reinforcement learning (with AlphaGo as an example) and wants to apply it in the domain of generative AI.
It’s worth noting that Gemini isn’t DeepMind’s first foray into language models. Last year, the company introduced Sparrow, a chatbot that the lab said was less likely than other models to give “unsafe” or “inappropriate” responses to questions.
Gemini is, however, DeepMind’s most ambitious work to date and gossips say that its development is accelerating Given the problems of Google Bard (another chatbot) to keep up with the pace imposed by ChatGPT and its massive use with which it continues to expand its neural network.
