Gemini Introduce: Comparing Google's AI to the GPT-4 Benchmark

Delhi : Google released Gemini, its "most capable" AI model, earlier this week, taking a huge step forward in the continuing quest for AI dominance versus OpenAI's GPT-4 and Meta's Llama 2.

Gemini was built from the ground up with 'multimodality,' allowing it to interpret and operate with several sorts of data at the same time, including text, code, audio, photos, and video.

Different mode option for Google Gemini

The AI model will come in three sizes: Ultra (for very complicated activities), Pro (for scalability over a wide variety of jobs), and Nano (for on-device tasks).

Following the merging of the search giant's AI research teams, DeepMind and Google Brain, this is the first model from Google DeepMind's stable.

After its late-year introduction, OpenAI's ChatGPT, powered by GPT-3.5, sent shockwaves around the world and rapidly became the talk of the town. While Google was taken off guard at first, it now appears to be getting its hands filthy and ready for a battle. Gemini outperforms GPT-4 on 30 of the 32 conventional performance criteria, according to Google DeepMind, albeit it's crucial to remember that the differences are small. While the firm has effectively presented the public with a 'Jetsons' fantasy, doubts over its authenticity have now taken main stage.

Benchmark comparison of Google Gemini against GPT-4

Google ran many benchmark tests that contrasted GPT-4 with Gemini. On the Massive Multitask Language Understanding (MMLU) exam, Gemini scored an astounding 90 percent, outperforming both GPT-4 (86.4 percent) and human experts (89.8 percent). A total of 57 courses, including maths, physics, history, law, medicine, and ethics, are used by MMLU to assess students' problem-solving and general understanding of the world.