Google took the wraps off its prolonged Deep Mind project now known as Gemini, a multimodal general artificial intelligence that will be flexible enough to power services and hardware from “data centers to mobile devices.”
Gemini is the brainchild of multiple internal Google businesses coming together such as Google Research, Google Search and Google’s infamous Deep Mind organization with the express intent to develop an AI that can seamlessly vacillate between generalized and specific types
Gemini version 1.0 is optimized through three distinct model sizes that include:
- Gemini Ultra — our largest and most capable model for highly complex tasks.
- Gemini Pro — our best model for scaling across a wide range of tasks.
- Gemini Nano — our most efficient model for on-device tasks.
Google’s Gemini comes on the heels of Microsoft preparing to infuse its own AI projects with wildly popular OpenAI’s latest platform update of GPT-4 and GPT-4 Turbo.
Comparably, Gemini has a lot going for it and may even begin to rival OpenAI and Microsoft’s co-dominance in the market with Google touting its “state-of-the-art score of 59.4 percent on a new multimodal benchmark test that translates into a higher score than ChatGPT for tasks across different domains usings deliberate reasoning.
With the image benchmarks we tested, Gemini Ultra outperformed previous state-of-the-art 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.
Sundar Pichai
CEO of Google and Alphabet
Demis Hassabis
CEO and Co-Founder, Google DeepMind
Gemini’s Ultra model notched a 90.0% score on Massive Multitask Language Understanding (MMLU) which measures world knowledge and problem solving across 57 subjects that include math, physics, history, law, medicine, ethics, and more. Gemini Ultra’s high score also denotes it outperforming human experts in the same MMLU measurement.
Developers should get a kick out of Gemini’s advanced understanding of coding. Once again, Google touts the exceeding performance of its Gemini Ultra model which garnered top benchmarks among HumanEval, and Natural2Code code tasks.
Beyond being a benchmark beast, Google is building Gemini to be more reliable, scalable and efficient as time goes on.
Today, we’re announcing the most powerful, efficient and scalable TPU system to date, Cloud TPU v5p, designed for training cutting-edge AI models. This next generation TPU will accelerate Gemini’s development and help developers and enterprise customers train large-scale generative AI models faster, allowing new products and capabilities to reach customers sooner.
Sundar Pichai
CEO of Google and Alphabet
Demis Hassabis
CEO and Co-Founder, Google DeepMind
Gemini 1.0 is being made available across a range of products and platforms starting with Google’s own Bard which is now accessible in over 170 countries and territories. Google plans to expand upon different modalities and supporting new languages and locations soon.
Gemini is also slowly being integrated into Google’s bread and butter Search business where it’s intended to improve the company’s Search Generative Experience (SGE) with a 40% latency reduction, at least in the US for now. Other experiences Gemin is planned to improve in the Google services lineup include Ads, Chrome and Duet AI in the future.
Google is also planning to bring Gemini to its Pixel lineup, specifically the Pixel 8 Pro, in the form of Gemini Nano to amp experiences like the Recorder app, Smart Reply in Gboard.
While Gemini Ultra did most of the heavy lifting in the benchmarks area, Google is still doing extensive testing and aims for early next year for access to select partners and developers in the form of Bard Advance, a new AI experience briefly mentioned.
