Google DeepMind, the cloud and tech giant’s AI research lab, is developing the LLM chatbot Gemini. It should be more powerful than competitor OpenAI’s ChatGPT. This writes Wired based on an interview with Google DeepMind CEO Demis Hassabis.
According to Demis Hassabis, Google DeepMind is going to vigorously compete with OpenAI. The main development that the AI research center is now working on is the chatbot Gemini.
This generative chatbot is based on the technology of DeepMind’s AI system AlphaGo. This AI system became the first AI solution to beat a human player in the board game Go in 2026.
Use reinforcement learning
Chatbot Gemini should not only be able to solve problems but also analyze text, Wired further writes. For this, it will combine the strength of AlphaGo systems with the language functionality of larger (LLM) AI models, Hassabis says.
According to Wired itself, Gemini will use new innovations in “reinforcement learning,” among other things. Reinforcement learning is a training of AI models where they are ‘rewarded’ for certain desirable behaviours and ‘punished’ for exhibiting undesirable behaviours. This should ultimately ‘teach’ the AI model what behaviour to exhibit in a given situation. Current (LLM) AI models still have difficulty dealing with this.
Previous project Sparrow
The development of Gemini is not Google DeepMind’s first project in this area. It previously worked on Sparrow. This chatbot is also less likely to give “unsafe” or “inappropriate” answers compared to other LLM-based chatbots.