“We’ve got to boldly and bravely go after those things.”
Gemini
According to Google DeepMind CEO Demis Hassibis, Google has an ultra-powerful large language model (LLM) on the way — and it may blow OpenAI’s existing systems like the ones powering ChatGPT out of the water.
In a new interview with Wired, Hassibis explained that the new AI dubbed Gemini is built on the capabilities of Google’s AlphaGo, a DeepMind-built machine learning system that famously became the first model to beat a human at the incredibly complex game Go back in 2016.
AlphaGo has long been considered a pioneer in the realm of reinforcement learning, a form of machine learning that deals with AI decision-making. By powering an LLM with AlphaGo’s deep decision-making skills, scientists hope it could one day overtake OpenAI’s buzzy products in tasks like reliable text analysis.
And, if Hassibis is to be believed, it could even put Google ahead in Silicon Valley’s ongoing and intensely competitive AI race.
“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,” Hassabis told Wired. “We also have some new innovations that are going to be pretty interesting.”
Heated Race
Before you get too excited, though, Hassibis did concede that Gemini is still very much under development and won’t make it to market for several more months.
Still, if Gemini does prove to possess superior problem-solving skills over OpenAI’s models — and Google still has a lot to prove — it could be a big deal.
For all of the strides OpenAI has made with its GPT-4 in recent months, in-context reasoning and decision-making are still where OpenAI’s latest iteration of its language model seems to falter.
As AI models are increasingly introduced into our online world, the ability for machines to reliably make sound decisions and correctly analyze text will only become more essential.
In any case, despite the widely-held view that Google has been playing catch-up to OpenAI in recent months, Hassibis seems confident about both Google and DeepMind’s powerful place in the AI landscape.
“If you look at where we are in AI, I would argue that 80 or 90 percent of the innovations come from [Google or DeepMind],” Hassabis told Wired.
“If done correctly, [AI] will be the most beneficial technology for humanity ever,” he added. “We’ve got to boldly and bravely go after those things.”
More on the AI race: Google Warns Employees about Using AI, While Promoting Its Own AI