What developers trying out Google Gemini should know about their data

Google says it may use data flowing through its Gemini API to improve its generative AI models, including those that power Google AI Studio and Gemini Pro.
Written by Eileen Yu, Senior Contributing Editor
Google Gemini
Omar Marques/SOPA Images/LightRocket via Getty Images

Developers who have jumped in to try out Google Gemini for free should know their data might be used to train its generative artificial intelligence (AI) models, including those that power Google AI Studio and Gemini Pro

The tech giant last week made Gemini Pro available to developers and businesses that are keen to build their own applications using its generative AI model. Developers can access the model via the Gemini API in Google AI Studio, while organizations will have to do so via Google Cloud's machine learning and development platform, Vertex AI

Also: AI will change the role of developers forever, but leaders say that's good news

Developers currently have free access to Gemini Pro and Gemini Pro Vision, capped at 60 requests per minute, which Google said is suitable for most app development requirements. The Gemini Pro Vision model allows text and imagery to be accepted as input, although output remains as text. 

Vertex AI developers can trial both AI models, within the same cap, for free until general availability, which is expected to be early 2024.

Also: How to use ChatGPT to write code

Following this date, charges per 1,000 characters or per image will apply across Google AI Studio and Vertex AI. Google said it had cut prices fourfold on input and twofold on output. 

Gemini Pro supports 38 languages and is available across more than 180 markets, including the Asia-Pacific region. 

Developers can move their AI Studio code to Vertex AI if they want a fully managed AI platform that offers more customization and Google Cloud features, including data governance and compliance, and security.  

Google, though, is touting AI Studio as the fastest way to build using Gemini.

Developers should note that when they use the free quota of 60 requests per minute, their API and Google AI Studio input and output "may be accessible to trained reviewers". 

Also: Generative AI means more productivity, and a likely retrenchment for software developers

Google told ZDNET that it uses the API inputs and outputs to improve product quality. "Human review is a necessary step of the model improvement process," a spokesperson said. 

"Through review and annotation, trained reviewers help enable quality improvements of generative machine-learning models like the ones that power Google AI Studio and the Gemini Pro via the Gemini API."

To protect developers' privacy, Google said their data is de-identified and disassociated from their API key and Google account, which is needed to log in to Google AI Studio. This protection takes place done before the reviewers can see or annotate the data. 

Also: AI in 2023: A year of breakthroughs that left no human thing unchanged

Google's Terms of Service (ToS) for its generative AI APIs further states that the data is used to "tune models" and may be retained in connection to the user's tuned models "[for] re-tuning when supported models change". 

The ToS states: "When you delete a tuned model, the related tuning data is also deleted." The terms also state that users should not submit sensitive, confidential, or personal data to the AI models. 

Data generated from when developers use Gemini Pro via Google AI Studio might still be accessed by Google reviewers, even if the developers make the move to Vertex AI. 

The data generated while users were on Google AI Studio will be tapped to help improve products, the Google spokesperson told ZDNET. 

"This includes further model tuning and evaluations. We may also derive product insights from anonymized data to help us determine new features we want to explore adding to Google AI Studio," they said.

Also: These 5 major tech advances of 2023 were the biggest game-changers

Developers and organizations with concerns about data security, but who are still keen to build with Gemini, will probably want to do so as Google Cloud customers, as this route will give them access to Vertex AI. 

Google has assured that this pathway provides "customization of Gemini with full data control". Accessing Gemini models via Vertex Ai also allows enterprise customers to tune the models with their own data. 

In addition, Google says it does not train its generative AI models on inputs or outputs from its cloud customers. 

Editorial standards