Definition:
Gemini is a family of artificial intelligence models developed by Google DeepMind. It is also the name of Google’s assistant, which uses those models and other tools to answer questions, create content and help with a range of tasks.
Its multimodal nature makes it possible to work with information in formats such as text, images, audio or video. The available capabilities depend on the model and service being used: the family of models and the Gemini application are related, but they are not the same thing.
Table of contents
The origins of Gemini and its relationship with Bard
Google introduced the Gemini family of models in December 2023 as part of its development of multimodal artificial intelligence systems.
Bard was the conversational assistant that Google had made available to the public that same year. In February 2024, it was renamed Gemini, so the assistant and the family of models began sharing the same name.
This was therefore not simply a matter of renaming a model called Bard. The evolution of the models and that of the assistant are related but distinct processes.
How Gemini works
Gemini generates responses based on its training, the instructions it receives and the available context. This may include the conversation, documents provided by the user or information obtained through connected tools.
The instructions, called prompts, specify the objective and expected format. For example, a person can ask for a document to be summarised, its technical terms explained or two sections compared.
Multimodality extends this interaction: a written question can be combined with an image or a supported file. Being able to interpret a format does not necessarily mean being able to generate it. Some assistant features use specialised models or tools to produce content.
What Gemini is used for
Gemini can support tasks involving the understanding, creation and organisation of information. Its applications include the following:
- Writing and editing: preparing drafts, summarising text, translating or adapting the tone of a message.
- Document analysis: locating information, comparing content and extracting data.
- Image interpretation: explaining screenshots, diagrams or other visual materials.
- Programming: explaining code, proposing changes and helping investigate errors.
- Learning: clarifying concepts, developing examples and preparing exercises.
- Support within other applications: using features integrated into services such as Gmail or Google Docs, depending on the account and configuration.
These tasks can be performed through the assistant or incorporated into a custom application that uses the models.
Ways to use Gemini
The Gemini assistant allows conversational interaction through its available interfaces. Some of its features are also incorporated into other Google products.
Developers can access the models through an API and integrate them into websites, applications or their own processes. In that case, the application determines what information is sent, which tools are involved and how the result is presented.
Integrations with documents and other services require the appropriate permissions. Using Gemini does not automatically grant access to all the information in a Google account.
Limitations to be aware of
Gemini can generate incorrect responses, misinterpret a file or cite information that does not support its explanation. Having search tools available does not mean that every response is up to date or verified.
Personalisation does not necessarily amount to retraining the model: using instructions, context or stored information involves different mechanisms.
Before providing confidential data, it is important to review the terms of the service being used. The assistant, enterprise features and developer integrations may have different data-handling policies.
The difference between Gemini and Google AI Studio
Gemini identifies Google’s family of models and its assistant. Google AI Studio is an environment for experimenting with models, preparing instructions and developing applications that use their capabilities.
A person can use the Gemini assistant to carry out a task directly. In AI Studio, they can test how a model should respond within an application they are preparing.
They are related tools, but serve different purposes: using Gemini does not require working in AI Studio.