According to Demis Hassabis, CEO of Google DeepMind, Google is currently building an AI system that will outperform ChatGPT.

Hasabis told Wired that the system, called "Gemini," is still under development and will take several months to complete. He also said that the cost of Gemini could run into the hundreds of millions of dollars.

Gemini runs on text and is expected to resemble ChatGPT-4 in many ways.

However, Google does not want to just replicate what it already has; according to the CEO, the company's engineers are using technology from AlphaGo, the AI that successfully defeated the champion of the Go board game, to power Gemini. This should give Gemini the ability to plan and solve problems. According to Hassabis, large-scale language modeling capabilities will be combined with these abilities.

Alpha Go was created using reinforcement learning. Reinforcement learning is when software repeatedly tries to solve a task and makes adjustments based on feedback about its performance.

Hasabis and his team may also try to enhance Gemini with ideas from other areas of AI, such as robotics and neuroscience.

"There are some new innovations that are going to be quite interesting," said Hassabis.

Gemini was announced at Google's I/O conference this year; Gemini is multimodal, has very efficient integration of tools and APIs, and is built to enable future innovations such as memory and planning, It was built from the ground up to achieve these goals. Although still in its early stages, we are already seeing impressive multimodal capabilities not seen in previous models."

He added that Gemini will be available in a variety of sizes and features.

In April, an internal document written by a Google engineer was leaked, stating that the company was losing its advantage in AI to the open source community, which can make quick adjustments to lean products. A month later, Google removed its chatbot Bard from its waiting list, Open AI's answer to ChatGPT. However, privacy concerns recently forced the company to postpone its European launch.

Bard made a factual error while answering questions in the first demo, costing parent company Alphabet $100 billion in market value.