Google has revealed a new AI model it claims beats rivals like ChatGPT at most tasks after the company’s “largest science and engineering project ever”.
Gemini is described as multimodal, meaning it can understand text, audio, video, images, and computer code.
It’s been built by teams across Google, including London-based AI research unit DeepMind.
DeepMind chief executive Demis Hassabis hailed it as “a significant milestone in the development of AI”.
Gemini has three levels of capability and the company said the highest ultra version, designed to carry out “highly complex tasks”, outperforms the likes of OpenAI’s GPT models in 30 of 32 academic benchmarks widely used in AI research and development.
A promotional video showed it assessing and giving feedback on a student’s physics homework.
It will not be released to the public until early next year.
The second tier is pro, which is already being rolled into Google’s Bard chatbot. It’s launching in more than 170 countries including the US, but has not yet been approved by regulators in the UK and EU.
Google said the upgrade will make Bard “more capable at things like understanding and summarising, reasoning, brainstorming, writing, and planning”.
Gemini’s third level is nano, which is designed to run locally on devices like Google’s Pixel smartphones. It will do things like summarise voice recordings and suggest replies in messaging apps.
The unveiling of Gemini is 2023’s latest major development in AI.
It comes 10 months after the release of Bard, which was widely reported to have been fast-tracked after the search giant was left spooked by the success of OpenAI’s ChatGPT in November 2022.
Wall Street welcomed the company’s bullish announcement of Gemini, with shares in parent firm Alphabet up 5%.
It’s also the first major release of a new AI model since the UK hosted a world-first safety summit last month, where Prime Minister Rishi Sunak announced tech companies had agreed to share data from their internal testing with the British government.
They are also required to share test data with the US government after President Joe Biden announced sweeping regulations designed to address concerns around AI.
The EU is expected to unveil its own regulatory proposals soon, with talks taking place this week.
According to the Associated Press, representatives from the bloc’s 27 member states and European Commission officials wrangled over details of the AI Act for 22 straight hours before pausing on Thursday.
Sky News has contacted Google for comment about their plans to share Gemini test data, and the reasons behind Bard’s upgrade being delayed in the UK and EU.
