[Courtesy of Google DeepMind]

<이미지를 클릭하시면 크게 보실 수 있습니다>

Global tech giant Google LLC released Gemini, its new generative artificial intelligence (AI) model, on Tuesday local time in a move to expand its presence in the intensifying AI battle.

Google held an online press conference for global media representatives on Tuesday and unveiled Gemini’s performance as well as various other indicators.

Gemini is a multimodal AI that can recognize images, speak, listen to voices, and even code. It also possesses advanced reasoning abilities to solve mathematical problems and analyze data.

Gemini was integrated into Google’s chatbot, Bard, on Wednesday.

Gemini scored 90 percent in massive multitask language understanding (MMLU), surpassing the human expert score of 89.8 percent, according to Google DeepMind Chief Executive Officer (CEO) Demis Hassabis.

He also added that Gemini topped the current best model’s performance in 30 out of 32 benchmarks, emphasizing that Gemini is superior to OpenAI’s GPT-4, which is currently rated as the best with a performance of 87 percent.

Google CEO Sundar Pichai said that the first version, Gemini 1.0, realizes Google DeepMind’s vision for the first time, describing it as the biggest scientific engineering achievement Google has undertaken as a company.

In the demo video released by Google, Gemini explained moves demonstrated by humans in real-time, reasoned through difficult problems, and even quizzed users.

Communication with Gemini was done entirely through speech as opposed to text, and Gemini translated the conversations into different languages as needed.

Gemini is presented in three models – Gemini Ultra, Pro, and Nano – based on performance and size.

The general-purpose Gemini Pro has been directly applied to Bard while the largest and highest-performing Gemini Ultra will be equipped on Bard under the name Bard Advanced in early 2024.