Developers, do you want to try Google Gemini? Be careful with your data

Picture: Omar Marques/SOPA Images/LightRocket via Getty Images.

Developers who use Google Gemini for free should know that their data can be used to train its generative artificial intelligence models, including those that power Google AI Studio and Gemini Pro.

Last week, Google made Gemini Pro available to developers and companies willing to create their own applications using its generative AI model. Developers can access the model via the Gemini API in Google AI Studio, while companies will have to do so via Google Cloud’s development and machine learning platform, Vertex AI.

A free and limited test version

Developers currently have free access to Gemini Pro and Gemini Pro Vision capped at 60 requests per minute. Which, according to Google, is suitable for most application development needs. With the Gemini Pro Vision model, requests can contain text or images, but the AI response will be in the form of text.

Developers using Vertex AI have the opportunity to test these two AI models for free, on a limited basis, until their general availability date – scheduled for early 2024. After this date, fees will apply to Google AI Studio and Vertex AI, per 1,000 characters or per image. Google announced that it has divided the prices by four for input data and by two for output data.

Currently, Gemini Pro supports 38 languages and is available in 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, as well as security. However, Google presents AI Studio as the fastest way to build with Gemini.

The data may be accessible to qualified Google employees

Developers should be aware that when they use the free quota of 60 requests per minute, their requests and the answers they receive “can be studied by qualified examiners”.

Indeed, as Google confirmed to ZDNET, the data is used to improve the quality of its products. “The review by a human being is a necessary step in the process of improving the models,” says a company spokesman. “With their analysis and feedback, qualified reviewers help improve the quality of generative machine learning models such as those that power Google AI Studio and Gemini Pro via the Gemini API. »

To nevertheless protect the privacy of developers, Google specifies that this data is depersonalized and disassociated from their API key and their Google account – a Google account is required to connect to AI Studio. This protection occurs before the examiners can see or annotate the data.

The data is kept to train the models

Google’s general terms of use (TOS) for its generative AI APIs also specify that the data is used to “adjust the models” and that it can be kept in connection with the models set by the user “for a new adjustment when the supported models change”.

“When you delete an adjusted model, the data related to it is also deleted,” reads the TOS. The latter also recommend not to transmit sensitive, confidential or personal data to its AI models.

In the event that a developer decides to switch to Vertex AI, the data he will have generated using Gemini Pro via AI Studio will still be able to be consulted by Google’s reviewers, this data being used to improve the brand’s products, explains a Google spokesperson to ZDNet. The improvements concern in particular “the development and evaluation of additional models”, he specifies. “From the anonymized data, it is also possible that we obtain information about our products to help us determine which features we would like to add to Google AI Studio. »

Google Cloud customer data is not reused

Developers and organizations concerned about the security of their data, but who still want to build with Gemini, should use the model through Vertex AI, as Google Cloud customers.

Google assures indeed that this way allows to “customize Gemini with full control of the data”. Access to Gemini models via Vertex AI also allows corporate clients to adjust the models with their own data.

In addition, Google claims that it does not train its generative AI models with data from queries from its cloud customers.

Source: ZDNet.com