Google DeepMind has unveiled its latest AI model, Gemini, which the company touts as its most advanced and versatile model to date. Gemini is a multimodal AI tool capable of comprehending and summarizing vast amounts of text, code, and audiovisual data. It comes in three sizes tailored for different tasks: Gemini Ultra, Gemini Pro, and Gemini Nano.
Gemini Ultra, the largest of the three models, is designed for complex tasks and has the ability to outperform humans in language comprehension testing. In fact, it achieved an impressive score of 90% in a massive multitask language understanding study, surpassing human capabilities across subjects such as physics, maths, history, and law.
In terms of language understanding and response generation, Gemini has proven its superiority over other AI models, even outperforming OpenAI’s GPT-4. Google’s CEO, Sundar Pichai, expressed his excitement about Gemini, highlighting its potential to make AI beneficial for people worldwide.
Gemini models are already integrated into Google’s products. For instance, Google’s large language model Bard now utilizes Gemini Pro, while users of the upcoming Pixel 8 Pro smartphone can expect new functionalities like auto-summarization, thanks to Gemini Nano.
Emphasizing their commitment to AI safety, Google asserted that Gemini models underwent the most comprehensive safety testing of any AI tool they have created thus far. Google’s in-house safety team conducted new research to mitigate potential risks associated with AI persuasion and autonomy. The company introduced a novel testing system and has made related papers available for reference.
The release of OpenAI’s ChatGPT in 2022 triggered a race among businesses across sectors to incorporate AI into their workflows. However, the challenge of accuracy has impeded complete adoption. Hallucinations, or the generation of factually incorrect responses, have hindered the potential use cases of AI. Accuracy remains a crucial hurdle in the development of AI models.
Research company GlobalData anticipates that AI models with 99% accuracy may be achievable within the next ten years, according to their second edition executive briefing on AI. It is worth noting that the timeline remains unpredictable since machine learning development fluctuates. Depending on various factors, it could take up to 30 years to reach the milestone of 99% accuracy.
In summary, Google DeepMind’s release of Gemini represents a significant stride in AI capabilities. With its powerful multimodal functionality and exceptional language comprehension, Gemini has the potential to revolutionize various industries. Google’s ongoing commitment to AI safety and their integration of Gemini models into their products further demonstrate their dedication to advancing the field of AI. As businesses continue to explore AI’s potential, achieving higher levels of accuracy remains a top priority, with progress expected within the next decade.
