Google CEO Sundar Pichai's Vision on AI
Via MIT Technology Review
The Story
In an interview tied to the launch of Gemini, Google and Alphabet CEO Sundar Pichai described the company's most capable AI model as natively multimodal from the ground up, built to process text, audio, images, and code at once.
He said Gemini Ultra reached state of the art results on 30 of the 32 leading benchmarks, excelling especially in multimodal tests.
Pichai highlighted that the model crossed the 90 percent threshold on the MMLU benchmark, becoming the first to do so after scores sat at 30 to 40 percent just two years earlier.
Gemini would reach users through Bard immediately and roll out across Google's products throughout 2024 in what he called one of the biggest upgrades yet.
He predicted AI would be one of the biggest things society grapples with over the next decade, bringing both surprising positive effects and real risks.
Pichai also acknowledged that hallucinations and data leakage remain unsolved research problems under current language model technology.
He said Gemini Ultra reached state of the art results on 30 of the 32 leading benchmarks, excelling especially in multimodal tests.
Pichai highlighted that the model crossed the 90 percent threshold on the MMLU benchmark, becoming the first to do so after scores sat at 30 to 40 percent just two years earlier.
Gemini would reach users through Bard immediately and roll out across Google's products throughout 2024 in what he called one of the biggest upgrades yet.
He predicted AI would be one of the biggest things society grapples with over the next decade, bringing both surprising positive effects and real risks.
Pichai also acknowledged that hallucinations and data leakage remain unsolved research problems under current language model technology.
Go Deeper
Read the original reporting at MIT Technology Review.
Read Full Story at MIT Technology Review →