Everything you need to know about Google Gemini

Google is one of the current tech giants working on generative artificial intelligence tools. Thanks to Anthropic and its Claude AI, it was one of the first responders to the launch of OpenAI’s now famous ChatGPT in late 2022.

However, it must be said that the first chatbot IA Bard, launched in February 2023, was not very convincing. Due to an error during a live demonstration, the AI somewhat tarnished the company’s reputation and even caused its stock market value to fall.

Advertising, continuation of your content below

But Google quickly caught up, launching Gemini in December 2023. Gemini is more than just a chatbot. It is a family of next-generation large language models (LLMs) capable of understanding natural language, images, audio, code, and video. Here’s everything you need to know about it.

Gemini (Google Bard)

The long-awaited Gemini conversational agent (Google Bard), powered by artificial intelligence and developed by Google, will be deployed this Thursday, July 13, 2023 in France as a free online service.

Downloads:
4041
Date of issue :
04/22/2024
Author:
Google
License:
Free license
Categories:

AI
Operating system :

Android, Internet service

From Bard to Gemini

Bard was Google’s first conversational AI, developed by the AI research labs DeepMind and Google Research. It was announced on February 6, 2023, and became available to users a few weeks later on March 21, 2023. At that time, you had to sign up for a waitlist to gain access to it. Google then removed this requirement and made its AI available in over 180 countries on May 10, 2023.

Bard was launched to compete with OpenAI’s ChatGPT, which was released in November 2022 and became an immediate hit. However, perhaps under pressure from the latter’s success, Google launched its AI before it was fully ready. During a public demonstration, the chatbot made a mistake by providing false information in response to a question asked by a user. A mistake that was not without consequences, as it cost Google a $100 billion stock market loss the next day.

Advertising, continuation of your content below

Of course, the company did not let itself be discouraged and continued to work on its AI. On December 6, 2023, Gemini was launched, a large language model with multimodal capabilities beyond those of models of the time (in this case GPT-4 from OpenAI).

The AI is touted to consist of a series of differently sized models, each designed for a specific set of use cases and deployment environments. The Gemini Ultra, the highest-end option, is designed to handle challenging tasks. Gemini Pro is the most versatile model and supports the Google Bard chatbot. Finally, Gemini Nano is designed for device use and is available on the Google Pixel 8 Pro.

Interestingly, Google’s AI chatbot is no longer called Bard as of February 8, 2024. It has been renamed Gemini, as have other Google AI tools integrated into some of its services (such as the former Duet tool for Google Workspace and Google Cloud).

What can you do with Gemini?

The three Gemini models were pre-trained on a variety of audio, photo and video materials, as well as a large set of codebases and text in different languages. Therefore, they are inherently multimodal and can reason about sequences of different types of inputs, including audio, images, and text.

Thanks to these capabilities, all Gemini models can theoretically answer questions in natural language, summarize and generate text, translate text into more than 100 languages, analyze complex visual elements (graphs, pictures, diagrams, etc.), subtitles and caption images. , translate audio, generate descriptions from videos, analyze and generate code (Python, Java, C++, etc.) and much more.

Advertising, continuation of your content below

One of Gemini’s strengths is its multimodal reasoning ability, meaning it can respond to a combination of different types of data in a single query.

Additionally, as stated above, each of the three Gemini models is designed for specific use cases:

Gemini Ultra, the ultimate version of LLM, is designed for very complex problems. According to Google, it can be used to help students with physics homework, to solve problems step-by-step, to identify scientific articles related to a particular problem, to extract information from an article, and much more.
Gemini Pro currently offered in version 1.5. Compared to its predecessor, it can handle more data, be it text (around 700,000 words or 30,000 lines of code), audio (up to 11 hours of audio) or video (up to an hour) in a wide range of languages. Remember, this is the LLM model behind Google’s Gemini AI chatbot.
Gemini Nano is a lightweight version of both models. Lightweight enough to run locally on some devices, it does not require sending a task to a remote server. As of this writing, it’s available on the Google Pixel 8 Pro, where it supports some AI features (including Smart Reply on the Gboard keyboard and Summarize in the Recorder app).

It depends on which Gemini model you want to access. If it’s the Gemini Nano, you can find it on the Google Pixel 8 Pro. However, it is expected to come to other devices in the future.

Otherwise, Gemini Pro is currently available in more than 230 countries and territories worldwide, including France. You can access it for free through the Gemini (formerly Bard) chatbot from your web browser. Just create a Google account.

Gemini Ultra is also quite widespread and is available in over 150 countries at the time of writing. However, access is paid. To use this model, you’ll need to subscribe to the Gemini Advanced option, which costs $20 per month. This can be done with a Google One AI Premium subscription, which also includes Google Workspace features and 2 TB of storage.

Finally, it’s worth noting that Gemini Pro and Gemini Ultra are also available in services such as Google Cloud Vertex AI (via API) and Google AI Studio.