Google has launched a new AI model called Gemini that it claims is the first of its kind to outperform humans in intelligence tests.
Gemini, launched on Wednesday, is “multimodal,” meaning it understands multiple types of data, including text, images and video unlike other AI models that are primarily trained on text, such as ChatGPT.
Google says Gemini is the “most capable general AI model” it has ever built, and is its largest science and engineering project ever.
Initial versions of Gemini called Nano and Pro began rolling out inside Google’s chatbot called Bard on Tuesday, and now also powers suggested messaging replies from the keyboard of Pixel 8 smartphones.
Gemini, which was developed by Google’s AI division called DeepMind, will be able to quickly summarize recordings made on Pixel 8 Pro and provide automatic replies on messaging services, starting with WhatsApp, Google says.
Google claims the more sophisticated Gemini Ultra model is the most human-like version to date, as it was able to outperform human experts in the industry-standard multitasking test called MMLU.
Gemini Ultra also outperforms current “state-of-the-art results,” including ChatGPT’s most powerful GPT-4 model, on 30 out of 32 academic benchmarks, Google says.
Tech expert Carmi Levy says the launch of Gemini is “more significant” than the launch of a new chatbot because it was built from the ground up as a fully multimodal model.
“When it’s engineered in from the beginning, and not simply grafted on later on, it is more efficient and effective – and smarter and more flexible. That means responses are more human-like, and its level of insight is significantly beyond existing LLMs,” Levy told Global News.
2:23 Google to pay $100M annually to news outlets as Canada’s Online News Act moves forward
LLM stands for Large Language Model, which is a type of AI that mimics human intelligence.
Levy says Google has “leapfrogged” OpenAI, which developed ChatGPT, meaning it can be expected that a Gemini-powered Bard is smarter.
Backed by Microsoft’s financial muscle and computing power, OpenAI was already deep into developing its most advanced AI model, GPT-4, when it released the free ChatGPT tool late last year. That AI-fuelled chatbot rocketed to global fame, bringing buzz to the commercial promise of generative AI and pressuring Google to push out Bard in response.
Can Canadians use Gemini?
Bard, which uses a fine-tuned version of Gemini Pro, is available in English in more than 170 countries and territories — but does not currently include Canada.
Google told Global News it is “enthusiastic about bringing Bard’s generative AI potential to Canadians soon,” but did not explain why it is not available at this time.
Levy says he believes Canadians can likely get their hands on Gemini sometime in the new year now that Google and the federal government have reached a deal over the Online News Act.
“This should open the door toward the company once again including Canada in global platform releases,” he said, noting that tensions around the act were likely to blame for the exclusion of Canada in Bard’s release around the world.
Heritage Minister Pascale St-Onge announced last week that Google will pay $100 million annually to Canadian news organizations as part of a new deal struck with Ottawa.
Google says Gemini can be useful in a number of industries and settings — one being in education. When presented with a physics problem including a hand-drawn slope and solutions written with a student’s messy handwriting, Gemini was able to understand the writing and convert both the problem and solution to mathematical typesetting.
“This opens up exciting educational possibilities, and we believe the new multimodal and reasoning capabilities of Gemini models have dramatic applications across many fields,” Google said.
1:46 U.S. Congress holds hearing on risks, regulation of AI: ‘Humanity has taken a back seat’
What are the concerns about Gemini?
However, Levy says that with any big release of an AI model like Gemini, there will be concerns about the risk of eclipsing human intelligence, or worse.
“The Gemini announcement will inevitably reignite debate over whether AI has gone too far, and whether we’re doing enough to prevent super-smart AI from damaging humanity,” he said.
Google has made attempts to ease anxiety around evolving AI by reassuring consumers that Gemini has been “rigorously tested.”
“At Google, there is this healthy disregard for the impossible, and that has oriented us to be both bold and responsible together,” said Google SVP of research, technology and society James Manyika in a Gemini introduction video.
“We develop proactive policies and adapt those to the unique considerations of multimodal capabilities. We then do rigorous testing against those policies to prevent the harms that we’ve identified,” said Tulsee Doshi, Google’s director of AI in the video.
Though Gemini Nano and Pro are already out, Ultra won’t debut until 2024, as it is pending “extensive trust and safety checks,” Google says.
It will be introduced into other Google products such as Chrome and its dominant search engine in the “coming months.”
Gemini’s launch is a “significant milestone” in the development of AI, Google says.
“We’ve made great progress on Gemini so far and we’re working hard to further extend its capabilities for future versions,” it says.
–with files from The Associated Press.News Related