Gemini, the newest AI model from Google, is available in three tiers: Nano, Pro, and Ultra. According to the company's internal sources, Gemini Ultra, the most potent of the three, beats OpenAI's GPT in practically all industry-standard metrics.
Gemini Ultra outperforms ChatGPT in 30 of the 32 commonly used academic benchmarks for assessing Large Language Models (LLM), according to industry insiders. This could help Google improve its situation, as the massive search engine has been accused of falling behind its AI rival OpenAI, whose GPT model is the most potent and well-liked in the sector.
As to an Insider article, Gemini Ultra achieved a score of 90% in the MMLU benchmark, making it the first model to surpass human specialists in this domain. With 57 subjects including math, physics, history, law, medicine, and ethics, MMLU assesses students' problem-solving skills as well as their general understanding of the world.
While Google plans to introduce Ultra in early 2024, its Gemini Nano and Pro AI versions are already available in the company's product line. Google's Pixel 8 Pro smartphone is equipped with AI features powered by Gemini Nano, and Gemini Pro may be accessible via the company's Bard chatbot. Through Google's Vertex AI initiative, enterprise clients and AI studio developers will also have access to Gemini Pro.
Eli Collins, the Vice President of Product at DeepMind, the Google unit in charge of creating the AI platform, claims that Gemini Ultra can comprehend "nuanced" data in a variety of media types, including text, graphics, audio, and code.
Collins disclosed that a portion of the information used to develop the application came from websites that were openly accessible. However, the corporation refrained from disclosing the precise sources that were utilized to train the AI.
