The CEO of Google’s DeepMind says that developers are building on technology from its 2016 AlphaGo model to outfit Gemini to be “more capable” than current LLMs.
Google’s artificial intelligence (AI) subsidiary DeepMind said the company’s forthcoming AI system will be “more capable” than OpenAI’s ChatGPT.
According to a report from Wired, Demis Hassabis, the CEO of DeepMind, says the company’s Gemini AI is a large language model (LLM) that utilizes text similar to ChatGPT’s processes but will be equipped with new capabilities like planning or problem-solving.
The system will be based on technology and techniques used in AlphaGo, an early AI system developed by DeepMind in 2016.
“At a high level, you can think of Gemini as combining some of the strengths of AlphaGo-type systems with the amazing language capabilities of the large models.”
He also mentioned that there will be “new innovations,” which he deemed “pretty interesting.” Gemini was first unveiled at Google’s I/O developer conference in May, along with a number of other AI products.
Cointelegraph reached out to DeepMind for further comment on the Gemini AI development.
Related: Google Cloud launches free courses to help users build their own GPT-style AI
Hassabis says Gemini is still months away from officially launching and will end up costing tens or hundreds of millions of dollars to develop.
He says he sees no reason why the things they’re building with Gemini “won’t work.”
In addition, the DeepMind CEO stressed the “urgent” need for further research into the likely risks of more capable AI systems. Previously Hassabis said that we could see human-level AI emerge before 2033.
The news of Gemini comes shortly after Google rushed out its own AI chatbot Bard, which rivals OpenAI’s hyped ChatGPT. However, Bard faced internal skepticism from former Google employees over “concerns for society.”
Already, Google Bard has been blocked from launching within the European Union due to regulators in Ireland accusing the company of failing to file the necessary paperwork.
Magazine: AI Eye: 25K traders bet on ChatGPT’s stock picks, AI sucks at dice throws, and more
