Google Cloud conducted an AI summit on December 14 and has announced Gemini, its most advanced large language model, accessible to businesses and developers via its API. Gemini is something that they described as being natively multimodal, having sophisticated reasoning skills, and advanced coding skills. The model comes in three variants: Ultra, Pro, and Nano.
As of yesterday, developers can access the Gemini Pro API through AI Studio, Google's no-cost, web-based development environment. Additionally, Gemini Pro is available for corporate use via Google Cloud's Vertex AI platform. This can be a huge opportunity for businesses to start creating applications immediately.
Google has announced plans to refine Gemini further in the upcoming weeks, guided by user feedback. The company expressed excitement in a recent blog post, eager to see the innovations developers and enterprises will create using Gemini.
What Is Gemini Pro Good For?
Gemini Pro on Vertex AI is a basically platform enabling developers to build AI applications, that are capable of processing text, code, images, and video, all at the same time. Let’s see some of the key features discussed during the summit:
- Gemini Pro is available alongside a selection of over 130 models from Google, open-source, and third-parties. These models are accessible as APIs for easy integration into applications.
- Developers can customize models with specific expertise, using tools to enhance training knowledge and adjust model weights. Vertex AI offers various tuning methods like prompt design, adapter-based tuning (e.g., LoRA), distillation, and reinforcement learning from human feedback (RLHF).
- Developers can use Gemini Pro to create AI agents quickly and efficiently. In the future, they plan that Gemini Pro will enhance search summarization and answer generation in Vertex AI, and be available as a foundation model for conversational agents.
- Google Cloud's data governance and privacy controls ensure customer data privacy, with Google not using customer data for model training.
Duet AI for Developers
Another very exciting news has been discussed during the summit: Duet AI for Developers has now been officially released. This tool is aimed at aiding developers in application creation and can be integrated within various Google Cloud interfaces. In the upcoming weeks, Gemini will also be incorporated into Duet AI.
Duet AI’s most remarkable features include:
- Generating code and unit tests
- Debugging code
- Answering questions about Google Cloud products
- Offering suggestions to better code and optimize
- Summarizing log entries
- Troubleshooting errors
Furthermore, the scope of Duet AI is extending to security operations. This collaboration tool will soon be available to security professionals within a SecOps platform. Within security operations, Duet AI will help in:
- Creating and running complex queries using only natural language
- Summarizing cases and adding context to help analysts triage
- Recommending further steps to teams so that they can act swiftly
It’s also important to mention that Duet AI for Developer will have no cost until February 1, 2024.
Conclusion
I’d like to conclude by quoting Thomas Kurian, the CEO of Google Cloud summarizing the latest announcements at the summit: “We brought Gemini to Vertex AI starting with Gemini Pro along with significant new features to build agents that combine native multimodality, advanced reasoning, up-to-the-minute knowledge and to take actions all finetuned to your data (…) We continue to take important steps to enable developers to build agents that think, understand, and act on information the way we as humans do. At the same time, we remain firmly committed to our promise to be bold and responsible in bringing you these advances.”
