Google unveils Gemini, its most capable AI model to date

Google CEO Sundar Pichai introduces Gemini, the company's most advanced AI model, with unparalleled performance across various modes of information.

Google CEO Sundar Pichai has announced the launch of Gemini, the company's most advanced AI model to date. This new system is designed to seamlessly understand and operate across various modes of information, including text, images, audio, and video.

Pichai expressed his excitement for the potential of Gemini, stating that it represents a significant science and engineering effort for the company. The model comes in three optimized sizes and has outperformed previous benchmarks, scoring higher than any other model on over 30 leading AI benchmarks.

Gemini Ultra, the largest version of the model, has achieved a major milestone by surpassing average human performance on a language test spanning over 50 academic subjects. With a score of 90.0%, Gemini Ultra is the first model to outperform human experts on massive multitask language understanding, which includes subjects such as math, physics, history, law, medicine, and ethics.

In addition to its language capabilities, Gemini has shown proficiency in computer programming. Google has utilized a specialized version of Gemini to create AlphaCode 2, an advanced code generation system that excels at solving competitive programming problems. AlphaCode 2 has demonstrated its ability by solving nearly twice as many problems as its predecessor and is estimated to outperform 85% of human participants.

Google has wasted no time in integrating Gemini into its products, including the Bard conversational AI tool and Pixel smartphones. Bard will now utilize a fine-tuned version of Gemini Pro for more advanced reasoning, planning, and understanding, marking the biggest upgrade to the tool since its launch. Additionally, Pixel 8 Pro has become the first phone enabled with Gemini Nano, powering new on-device features like text summarization in apps.

While Gemini Ultra is set to be available through Google's developer platforms in 2023, select developers and partners will have early access to the model later this year. Google is currently conducting safety checks on the model to ensure its reliability and security.