Google has announced the next step in its AI journey with Google Gemini, its most advanced model yet. This comes eight years into their journey as an AI-first company. Gemini 1.0, the first version, is optimized for different sizes: Ultra, Pro, and Nano. These models represent one of the biggest science and engineering efforts undertaken by Google. Pichai believes the transition to AI will be the most profound in our lifetimes, bringing new waves of innovation and economic progress. He also emphasized the importance of addressing risks as AI becomes more capable.
Google DeepMind has introduced Gemini, a new generation of AI models. Gemini is the result of collaborative efforts across Google and is designed to understand and operate across different types of information, including text, code, audio, image, and video. It can run efficiently on everything from data centers to mobile devices.
“AI has been the focus of my life’s work, as for many of my research colleagues. Ever since programming AI for computer games as a teenager, and throughout my years as a neuroscience researcher trying to understand the workings of the brain, I’ve always believed that if we could build smarter machines, we could harness them to benefit humanity in incredible ways.”
Demis Hassabis, CEO and Co-Founder of Google DeepMind
Introduction to Gemini: A New AI Model
Google DeepMind has introduced Gemini, a new generation of AI models. This model is designed to be more intuitive and useful, acting as an expert helper or assistant. Gemini is the result of large-scale collaborative efforts across Google, including Google Research. It has been built to be multimodal, meaning it can understand, operate across and combine different types of information including text, code, audio, image and video.
Gemini is also highly flexible, able to run efficiently on everything from data centres to mobile devices. This will significantly enhance the way developers and enterprise customers build and scale with AI. The first version of Gemini, Gemini 1.0, has been optimised for three different sizes.
Gemini’s Performance and Capabilities
Gemini has been rigorously tested on a wide variety of tasks. From natural image, audio and video understanding to mathematical reasoning, Gemini Ultra’s performance exceeds current results on 30 of the 32 widely-used academic benchmarks used in large language model (LLM) research and development. Gemini Ultra is the first model to outperform human experts on MMLU (massive multitask language understanding), which uses a combination of 57 subjects such as math, physics, history, law, medicine and ethics for testing both world knowledge and problem-solving abilities.
Gemini Ultra also achieves a high score on the new MMMU benchmark, which consists of multimodal tasks spanning different domains requiring deliberate reasoning. With the image benchmarks tested, Gemini Ultra outperformed previous models, without assistance from object character recognition (OCR) systems that extract text from images for further processing. These benchmarks highlight Gemini’s native multimodality and indicate early signs of Gemini’s more complex reasoning abilities.
Gemini’s Advanced Reasoning and Understanding
Gemini 1.0’s sophisticated multimodal reasoning capabilities can help make sense of complex written and visual information. This makes it uniquely skilled at uncovering knowledge that can be difficult to discern amid vast amounts of data. Its remarkable ability to extract insights from hundreds of thousands of documents through reading, filtering and understanding information will help deliver new breakthroughs at digital speeds in many fields from science to finance.
“Gemini is the result of large-scale collaborative efforts by teams across Google, including our colleagues at Google Research. It was built from the ground up to be multimodal, which means it can generalize and seamlessly understand, operate across and combine different types of information including text, code, audio, image and video.”
Demis Hassabis, CEO and Co-Founder of Google DeepMind
Gemini 1.0 was trained to recognise and understand text, images, audio and more at the same time, so it better understands nuanced information and can answer questions relating to complicated topics. This makes it especially good at explaining reasoning in complex subjects like math and physics.
Gemini’s Coding Capabilities
The first version of Gemini can understand, explain and generate high-quality code in the world’s most popular programming languages, like Python, Java, C++, and Go. Its ability to work across languages and reason about complex information makes it one of the leading foundation models for coding in the world. Gemini Ultra excels in several coding benchmarks, including HumanEval, an important industry-standard for evaluating performance on coding tasks, and Natural2Code, an internal held-out dataset, which uses author-generated sources instead of web-based information.
Using a specialised version of Gemini, a more advanced code generation system, AlphaCode 2, was created which excels at solving competitive programming problems that go beyond coding to involve complex math and theoretical computer science. When programmers collaborate with AlphaCode 2 by defining certain properties for the code samples to follow, it performs even better.
“Today, we’re a step closer to this vision as we introduce Gemini, the most capable and general model we’ve ever built.”
Demis Hassabis, CEO and Co-Founder of Google DeepMind
Responsibility and Safety in Gemini’s Development
Google is committed to advancing bold and responsible AI in everything they do. Building upon Google’s AI Principles and the robust safety policies across their products, new protections have been added to account for Gemini’s multimodal capabilities. At each stage of development, potential risks are considered and work is done to test and mitigate them.
Gemini has the most comprehensive safety evaluations of any Google AI model to date, including for bias and toxicity. To limit harm, dedicated safety classifiers have been built to identify, label and sort out content involving violence or negative stereotypes. This layered approach is designed to make Gemini safer and more inclusive for everyone. Responsibility and safety will always be central to the development and deployment of Google’s models.
“With a score of 90.0%, Gemini Ultra is the first model to outperform human experts on MMLU (massive multitask language understanding), which uses a combination of 57 subjects such as math, physics, history, law, medicine and ethics for testing both world knowledge and problem-solving abilities.”
Demis Hassabis, CEO and Co-Founder of Google DeepMind
“We designed Gemini to be natively multimodal, pre-trained from the start on different modalities. Then we fine-tuned it with additional multimodal data to further refine its effectiveness. This helps Gemini seamlessly understand and reason about all kinds of inputs from the ground up, far better than existing multimodal models — and its capabilities are state of the art in nearly every domain.”
Demis Hassabis, CEO and Co-Founder of Google DeepMind
Summary
Google DeepMind has introduced Gemini, a new generation of AI models that can understand and operate across different types of information including text, code, audio, image and video. Gemini’s advanced capabilities, such as complex reasoning and high-quality code generation, aim to enhance the way developers build and scale with AI, and its potential applications range from science to finance.
- Google DeepMind, led by CEO Demis Hassabis, has introduced Gemini, a new generation of AI models.
- Gemini is a result of collaborative efforts across Google and is designed to be multimodal, meaning it can understand and operate across different types of information including text, code, audio, image and video.
- Gemini is flexible and can run on various platforms, from data centres to mobile devices.
- The first version, Gemini 1.0, comes in three sizes: Gemini Ultra for complex tasks, Gemini Pro for a wide range of tasks, and Gemini Nano for on-device tasks.
- Gemini Ultra has outperformed human experts on MMLU (massive multitask language understanding), a test that uses a combination of 57 subjects for testing world knowledge and problem-solving abilities.
- Gemini is designed to be natively multimodal, pre-trained on different modalities and fine-tuned with additional multimodal data.
- Gemini 1.0 can understand, explain and generate high-quality code in popular programming languages like Python, Java, C++, and Go.
- Google has also introduced Cloud TPU v5p, a new TPU system designed for training AI models, which will accelerate Gemini’s development.
- Google has conducted comprehensive safety evaluations of Gemini, including for bias and toxicity, and is working with external experts to stress-test the models.
“At Google, we’re committed to advancing bold and responsible AI in everything we do. Building upon Google’s AI Principles and the robust safety policies across our products, we’re adding new protections to account for Gemini’s multimodal capabilities. At each stage of development, we’re considering potential risks and working to test and mitigate them.”
Demis Hassabis, CEO and Co-Founder of Google DeepMind
