Introduction:
Google’s Gemini is a suite of generative AI models, apps, and services developed by Google’s AI research labs DeepMind and Google Research. It consists of three models: Gemini Ultra, Gemini Pro, and Gemini Nano. These models are “natively multimodal,” meaning they can process more than just text and are pre-trained on various types of data such as audio, images, videos, codebases, and text in different languages. This sets Gemini apart from other models like Google’s LaMDA, which can only work with text data.
Gemini Apps vs. Gemini Models:
It’s important to note that Gemini is separate from the Gemini apps available on the web and mobile. The Gemini apps serve as an interface to access specific Gemini models. Google’s branding strategy may have caused some confusion initially, but it’s crucial to understand the distinction between Gemini as a model family and the apps that enable access to those models. Additionally, Gemini models are independent of Google’s Imagen 2, a text-to-image model available in some of the company’s development tools.
Capabilities of Gemini Models:
Due to their multimodal nature, Gemini models have the potential to perform a wide range of tasks, including transcribing speech, captioning images and videos, and generating artwork. While some of these capabilities are still in development, Google promises that they will be available in the future. However, it’s essential to approach these claims with caution, considering Google’s history of underdelivering and the recent controversy surrounding a doctored video showcasing Gemini’s abilities.
Gemini Ultra:
Gemini Ultra is the most performant model in the Gemini family. It can assist with tasks like physics homework, problem-solving on worksheets, and identifying relevant scientific papers. Additionally, it has the potential to generate images directly without an intermediary step. Gemini Ultra is available through Google’s Vertex AI and AI Studio platforms but requires a subscription to the Google One AI Premium Plan, priced at $20 per month. It can also integrate with Google Workspace, making it useful for summarizing emails or capturing notes during video calls.
Gemini Pro:
Gemini Pro is an improvement over Google’s LaMDA in terms of reasoning, planning, and understanding capabilities. A study by Carnegie Mellon and BerriAI researchers found that the initial version of Gemini Pro outperformed OpenAI’s GPT-3.5 in handling longer and more complex reasoning chains. However, the study also highlighted some limitations, such as difficulties with mathematics problems involving multiple digits. Google addressed these issues with the release of Gemini 1.5 Pro, which can process significantly more data and analyze audio and video content in multiple languages.
Gemini Nano:
Gemini Nano is a smaller version of the Gemini Pro and Ultra models, designed to run directly on mobile devices like the Pixel 8 Pro and Samsung Galaxy S24. It powers features like Summarize in Recorder and Smart Reply in Gboard. The Recorder app provides Gemini-powered summaries of recorded audio, ensuring privacy as no data leaves the user’s phone. Smart Reply suggests responses in messaging apps and will expand to other apps over time.
Gemini vs. OpenAI’s GPT-4:
Google claims that Gemini outperforms OpenAI’s GPT-4 on several benchmarks, but the differences are marginal. Users and academics have reported issues with Gemini Pro, including incorrect facts, translation difficulties, and poor coding suggestions. While benchmarks can provide some insight, they may not fully reflect the model’s performance in real-world scenarios.
Cost of Using Gemini:
Gemini 1.5 Pro is currently free to use in the Gemini apps, AI Studio, and Vertex AI during the preview period. However, once it exits preview, there will be costs associated with its usage. Gemini 1.5 Pro will be priced at $0.0025 per character for processing and $0.00005 per character for output. Pricing for Gemini Ultra has not been announced yet.
Availability of Gemini:
Gemini Pro and Ultra can be accessed through the Gemini apps, Vertex AI, and AI Studio. The Gemini apps are the easiest way to experience Gemini, while Vertex AI provides API access. AI Studio offers workflows for creating structured chat prompts and customization options. Gemini models are also integrated into Google’s development tools, including Chrome, Firebase, and database creation and management tools. Additionally, Gemini powers security products like Gemini in Threat Intelligence, enhancing Google’s cybersecurity platform.
Conclusion:
Google’s Gemini is an ambitious suite of generative AI models that aims to provide multimodal capabilities for various tasks. While the models show promise, there have been concerns raised about their performance and limitations. Users should approach Gemini’s capabilities with caution and consider factors like cost and specific use cases when deciding to incorporate these models into their projects.
