In a groundbreaking announcement, Google has introduced Gemini, its latest and most advanced AI model. Developed through extensive collaboration across Google teams, including Google Research, Google Gemini, according to the company, sets a new standard for multimodal capabilities, seamlessly combining and understanding diverse data types such as text, code, audio, image, and video.
Three Versions of Google Gemini
Apart from being a capable AI model, Gemini is also the most flexible, designed to operate efficiently across various platforms—from data centers to mobile devices, says Google. The model comes in three optimized versions:
- Gemini Ultra, tailored for highly complex tasks.
- Gemini Pro, ideal for scaling across a wide range of applications.
- Gemini Nano, the most efficient model for on-device tasks.
How Google Gemini Could be Better than ChatGPT
In rigorous testing, Gemini Ultra 1.0 has demonstrated unparalleled performance, surpassing human experts in Massive Multitask Language Understanding (MMLU) with an impressive score of 90.0%, said the company in its official blog post. This benchmark, encompassing 57 subjects ranging from math and physics to history and medicine, showcased Gemini's ability to combine world knowledge and problem-solving skills. Gemini's approach to MMLU allows the model to engage in thoughtful reasoning before providing answers to challenging questions, leading to significant improvements over traditional models, added Google indicating why it may be better than ChatGPT.
In addition to that, Gemini 1.0’s multimodal reasoning capabilities make it uniquely adept at deciphering complex written and visual information. Its prowess in extracting insights from vast datasets is poised to accelerate breakthroughs across various fields, from science to finance, says Google. With the ability to read, filter, and understand information across hundreds of thousands of documents, Gemini promises to deliver unprecedented advancements at digital speeds. The era of Gemini marks a pivotal moment in AI development, offering a tool that can navigate and derive meaningful insights from the ever-expanding sea of information in our interconnected world, according to Google.
How Google Gemini Aims at Solving AI Concerns Such as Bias?
In a commitment to advancing responsible AI practices, Google is fortifying safety measures to align with the multimodal capabilities of its latest AI model, Gemini. Gemini is undergoing the most comprehensive safety evaluations ever conducted for a Google AI model, encompassing assessments for bias and toxicity, says the company. Novel research has been undertaken to explore potential risks in areas such as cyber-offense, persuasion, and autonomy. Leveraging Google Research’s adversarial testing techniques, critical safety issues are proactively identified prior to Gemini's deployment.
To ensure a thorough evaluation, Google is also collaborating with a diverse group of external experts and partners to stress-test Gemini across a spectrum of issues, addressing potential blind spots in the internal evaluation approach. In diagnosing content safety issues during Gemini's training phase, Google says that it is utilizing benchmarks like Real Toxicity Prompts. This benchmark, comprising 100,000 prompts with varying degrees of toxicity sourced from the web, has been developed by experts at the Allen Institute for AI.