4 min Devops

What should Google’s Gemini offer developers?

What should Google’s Gemini offer developers?

Google is bringing Gemini’s capabilities to the places developers already are. That brings the possibility to personalize and fine-tune Gemini to the wishes of their organizations. For Europe, however, the most important news seems to be that the Imagen AI image generator is getting a second version in Vertex AI.

Google announced it is making the Gemini API available to developers. The AI model will come to various Google environments, where developers can get started with the capabilities of this “next-generation AI model”.

Behind the Gemini API is Gemini Pro. Gemini Ultra, the most powerful version of the newly available AI model, is still in the testing phase. With the most scalable Gemini version, developers should already get more accurate answers. To that end, Thomas Kurian, CEO of Google Cloud, insists that Gemini is trained to handle different input types.

It is, therefore, a setback that this first version of the API has limits on the types of data sources that can be processed. Only text and images are accepted for input, while output is only available in text form. In order to load images, it is important to use Gemini Pro Vision. It is clear that with Gemini Pro, developers are not yet getting quite what Google promised with the launch of Gemini. While the capabilities for audio, video and code should make the AI model unique.

Also read: Google changes its mind and lets Gemini compete with GPT-4 immediately

In Vertex AI and Google AI Studio

In the demonstration for Gemini in Vertex AI, we see text combined with images to make the prompt richer in information. Nenshad Bardoliwalla, Director of Product Management at Vertex AI, shows how the model can help a housing rental agent. This user can ask the chatbot to create a description based on a text prompt and some images of the property. This allows the chatbot to better respond to what the property really looks like, whereas previously, this was not possible unless the landlord wrote all this information themselves in the text prompt.

Furthermore, it shows the ability to fine-tune the model. So, suppose this property is from a professional renting company; in that case, the text should follow a certain house style. For that, developers in Vertex AI can customize Gemini Pro to their liking. That fine-tuning is also possible by attaching publicly available information to the model. For example, a website that rents out houses can be attached to the model so that Gemini can, for example, determine what normal prices are for the next three weeks (including the Christmas period) based on what the other owners on the website are offering for rent.

Developers can further work with the models in several popular programming languages. Specifically, Python, Kotlin (Android programming language), Node.js, Swift and JavaScript are supported.

Gemini Pro will be available immediately in Google AI Studio and Vertex AI. Google wants to encourage the use of Gemini with a free limit set. That assumes 60 free requests per minute. That is a limit 20 times higher than what other AI model providers offer for free, according to the company.

Imagen 2 for Vertex AI

The cloud company has more announcements in store for Vertex AI. There, the second version of AI generator Imagen is now becoming available. That model gives users more control over text in the image. For example, Kurian showed that it is possible to place a logo on a tube of toothpaste in the AI-generated image.

There are also improvements in the quality of the generated images. For now, Imagen 2 is available in seven languages, including English, Portuguese and Spanish. More languages should be added to the list in the spring of 2024.

Europe lags behind

Similar to Bard’s launch with Gemini Pro, Gemini will not yet be available to European developers. Kurian could say very little about availability in Europe: “Gemini is available to some European Google customers, and they will also have access to other features.” That doesn’t tell us anything more about when Gemini will then become generally available in Europe.

The same goes for the availability of the model in Duet AI, the AI assistant for Workspace users. Google announced two new specialized assistants there today. They are Duet AI for developers and Duet AI for security operations. Developers can expect help in all areas of the development phase. “It helps write code, deploy applications, scale them and help diagnose problems once the app is working,” says Gerald.

Also read: Gemini vs. GPT-4: Google shows what a next-gen AI model offers