The man in charge of Google’s DeepMind AI effort has made a rare admission that the company was caught napping when OpenAI blitzed the market with ChatGPT.
Demis Hassabis, co-founder and CEO of Google DeepMind, the company’s cutting edge AI operation, said Google hadn’t anticipated that OpenAI would go to market so early without first ironing out ChatGPT’s defects. But the public were keen to embrace the new tool, no matter what.
Dr Hassabis made the admission during an enthralling discussion (see separate report) on the application of AI to modifying proteins in the body to fight disease. He spoke before a packed audience at a seminar on the future of AI at Mobile World Congress in Barcelona.
Google’s tardiness in getting its Gemini generative AI product to market stunned many observers who for years saw Google as being way ahead in the AI product race.
In 2016 the Google Brain Team had released its open-sourced TensorFlow hardware designed to accelerate machine learning. That project involved more than 480 people and it seemed that Google was riding high in the AI stakes.
Google had been developing its AI offering through at least two entities: Google Brain and Google DeepMind. Google Brian was Google’s own AI research division formed in 2011. DeepMind was an AI research laboratory founded in the UK in 2010 and acquired by Google in 2014. The two coexisted until last year.
Dr Hassabis said Google had built a formidable reputation at the research cutting edge. “We have this sort of long, storied history of doing the innovations and doing the basic science. We’re very, very good at that,” he said.
“But what OpenAI did, and I give kudos to them, was that they took these ideas and applied the Silicon Valley growth mentality to hacker mentality and scaled it to a maximum speed. I don’t think anyone really predicted, maybe even including them, that these new capabilities would just emerge, just through scaling – not from inventing some new innovation, but actually just sort of scaling.
“And it’s quite unusual in the history of most scientific technology fields where you get step changing capability by doing the same thing just bigger. That doesn’t happen very often.
“Usually you just get incremental capabilities. And normally you have to have some new insight or some new flash of inspiration, or some new breakthrough in order to get a step change. And that wasn’t the case here.
“So when ChatGPT came out, the other surprising thing that I thought even surprised Open AI, was that the general public seemed to be really ready to use these systems, even though they clearly have flaws – hallucinations, they’re not factual sometimes …
“We were sort of thinking, maybe they’re not ready for primetime. They need to get them down to 100x more accurate and other things, but obviously, OpenAI just put it out there, and it turns out billions of people found value out of that for gaming, today’s use cases: coding, summarizing documents and so on.
“And it didn’t have to be 100 percent accurate for there to be some valuable use case. So I think that was the surprising thing for the whole industry. And I would include OpenAi in that. Clearly the general public were ready to use these things perhaps a couple of years ahead of when we were expecting.”
History shows that Google suffered by not being first-to-market. Google Brain and DeepMind were merged in April 2023. By mid 2023, co-founder Sergey Brin had been brought back to get Google’s Gemini generate AI project back on track.
Brin was working three to four days at Google’s campus, reported The Wall Street Journal. It was a marked contrast to his hands-off approach after stepping down in 2019.
And life would remain unpredictable for Google staff, with the company announcing two rounds of job cuts at the start of 2024 and Alphabet and Google CEO Sundar Pichai warning of more job cuts on the way, as the company went onto a war footing to concentrate on getting back ahead in the AI race.
