DeepMind, the brainy folks building AI products at Google, is developing a new chatbot called Gemini to take on OpenAI’s ChatGPT.
Gemini is a large language model, similar to GPT-4, and it’s currently under development.
It’s anticipated to cost tens to hundreds of millions of dollars, comparable to the cost of developing GPT-4.
To make Gemini good at problem-solving, they’re borrowing “a reinforcement learning technique” from AlphaGo, where the AI is rewarded for good behavior and corrected for mistakes. It’s like training your pet!
Backstory: AlphaGo made history by defeating a champion Go player in 2016 using reinforcement learning and tree search methods. These techniques are planned to be used in Gemini.