[For Engineers] Gemini Usage Guide

Hello, this is the official note editorial department for Google's AI, "Gemini."
In this article, we will focus on the use of Gemini in corporate settings and introduce ways for engineers to utilize it.

In development environments, the use of generative AI has become a viable option. In this era of rapidly evolving generative AI,"I want to develop a system using generative AI, but I don't know how to use it specifically," or "I've used other generative AIs, but how is Gemini different?"Many of you may be thinking things like this.

As a starting point to answer such questions, Etsuji Nakai, a Google Cloud employee, recently appeared on the PIVOT program "&TALK" to share three ways to use Gemini.These are the "Gemini app" for general users, "Gemini for Google Workspace" for business users, and the "Vertex AI Gemini API" for engineers.

Those who are not IT specialists can use the Gemini app conveniently to improve their daily lives. Business users can use Gemini for Google Workspace to streamline their daily business operations. And IT experts and IT engineers can make full use of the very brain of Gemini via Google Cloud's Vertex AI to create new services and systems. We shared that vision of the future.

In this article, to help you, the engineers, learn more about Gemini, we will delve deeper intohow to use Gemini for development.Please find the way to start with Gemini that suits your skill level and development environment.

Gemini is being used more and more in development; what is the reaction from the field?

―In the PIVOT program, you concluded by saying, "I want IT engineers to use Gemini." What kind of reactions are you getting from engineers regarding Gemini?

For those who have never touched generative AI, I recommend trying the Gemini app first, but the Gemini app is limited to a chat format, and it is often thought that this is all there is to Gemini. Therefore, there are many cases where people are desperately searching for ideas on "what can be done with chat-type generative AI."

However,the true value of Gemini lies in "being able to incorporate its brain into existing applications and services."I hope for an expansion of ideas, not just creating something new with generative AI, but utilizing Gemini to make already developed apps and services more attractive.

―There are already engineers who have successfully incorporated Gemini into their applications, right?

Engineers who are actually incorporating Gemini into their own applications understand the areas where Gemini excels and where it struggles. At the same time, they also understand the role their company's applications play in business, and I believe that it is only with thisunderstanding of both "generative AI technology" and "our own business"that the optimal combination becomes visible.

Especially in companies that operate their own services, engineers often understand the background of the service, making it easier to find effective ways to use Gemini. Regarding the unexpected output (hallucinations) unique to generative AI, they do not make a negative judgment that "it's unusable because of this," but rather view it from a constructive perspective of "it's acceptable if used this way." Not limited to generative AI, all technologies have their own inherent constraints. It is important to develop the ability to find the optimal way to use them while keeping business goals in mind.

―When using Gemini, what kind of communication should engineers have with the business side?

From an engineer's perspective, rather than getting too carried away with the possibilities that open up with the use of generative AI, I think it is good to be conscious ofquickly creating a simple prototype based on what is essentially important to the business and showing it to the business side first.

By having a concrete implementation example, it becomes easier for the business side to grasp the content of the development, and more focused discussions become possible.

3 Patterns of Gemini Usage: What are the suitable ways to use it for each development environment?

―When thinking about using Gemini from a developer's perspective, what are the criteria for choosing the optimal solution?

Gemini usage methods can be broadly classified into three patterns.
First, the Gemini app, which is turned into an application so that end users can use it immediately. Next, Gemini for Google Workspace, where Gemini is incorporated into Google Workspace, and finally, the Gemini API for engineers.

Among these, what I would like you, the engineers, to pay attention to is theGemini API.By using this, you canextract the brain part of Geminithat runs behind the Gemini app and Gemini for Google Workspace and incorporate it into various applications. In PIVOT's "&TALK," I introduced it as the "Vertex AI Gemini API," but besides Vertex AI, it can also be used from Google AI Studio.

―Should I use Google AI Studio or Vertex AI?

For engineers already using Google Cloud, the best choice is to utilize the Vertex AI Gemini API within that environment. You can combine various Google Cloud managed services with the Gemini API to accelerate system development. On the other hand, if you do not have a Google Cloud environment or simply want to try out the Gemini API first, you can use Google AI Studio.

Google AI Studio has both free and paid versions. The free version has limitations on available resources and may not be suitable for business use when considering corporate security requirements. Conversely, the paid version reduces errors caused by resource shortages and explicitly states that user data is not used for model training by Google, making it suitable for enterprise-level development and full-scale business applications.

―Compared to Google AI Studio, what are the advantages of Vertex AI for engineers?

The Gemini API provided by Google AI Studio and Vertex AI basically has the same functionality. However, since the Vertex AI Gemini API is integrated into the Google Cloud environment, it works very smoothly with other Google Cloud features.

The ultimate goal for engineers is to integrate the Gemini API into their own applications, which requires coordination with various components such as networks and databases. Since Vertex AI is self-contained within the Google Cloud environment, it has mechanisms in place to automate these integrations, enabling efficient development. Security and SLA settings tailored for enterprise-level use are also benefits of using Google Cloud.

―Which environment do you recommend for engineers who need to 'build a prototype'?

For engineers who develop on Google Cloud daily, developing in the Google Cloud environment where the necessary development resources are already available is smoother. As a result, demonstrations to the business side can also be conducted speedily. If you are only using the Gemini API, Google AI Studio is an easy-to-use free option, but it is important to remember that there are certain limitations on resource usage.

―As a recent development, Gemini can now be called from OpenAI libraries; what kind of changes will this bring to development sites, including for engineers?

Currently, there are many large language models (LLMs) like Gemini, each with its own characteristics. Compatibility with actual applications varies, and you won't know if it will perform as expected until you actually try it.

Even if you are developing with OpenAI products, there are many cases where switching the backend model to Gemini can solve issues that were not working well. Previously, such a switch required application modifications, but now that Gemini can be called directly from OpenAI libraries, it has become much easier to switch.

This creates an environment where engineers can easily test Gemini's performance, enabling more flexible development.

Points to note for engineers using Gemini

―Are there any points engineers should be careful about to master Gemini in development?

The key to mastering Gemini lies in how you write your prompts. While there is information on the internet suggesting that 'there are spell-like special prompts for LLMs,' such special techniques are not necessary for Gemini.

If you write your prompts so that 'what you want it to do' is clearly communicated, Gemini will work as expected. When you get a result different from what you expected, instead of giving up and thinking, 'Gemini is no good after all,' first look back and ask yourself if your instructions were ambiguous. Just like programming, to get the expected results, you need to logically organize and convey what you want done. The saying, 'Programs don't run as you think; they run as you write them,' also applies to prompts.

Of course, writing logical instructions like this requires experience. Prompt engineering skills steadily improve through practice.

Conclusion

In this article, we explained how to use Gemini for engineers. We covered the characteristics and usage points of each, from Google AI Studio, which you can try easily for free, to Vertex AI, which is a full-scale development environment.
Gemini is a tool that can incorporate new intelligence into existing applications and increase the value of businesses and services. Please try using Gemini in the best way suited to your development environment and goals.

Experience Google Gemini for free

Next time, as more practical content, we will explain the specific introduction procedures for Vertex AI and Google AI Studio. We will also introduce effective prompt techniques to maximize Gemini's capabilities.