Google Gemini (Gemini AI)

Ross Jukes
Last updated: May 27, 2024
Why Trust Us
Our editorial policy emphasizes accuracy, relevance, and impartiality, with content crafted by experts and rigorously reviewed by seasoned editors for top-notch reporting and publishing standards.
Purchases via our affiliate links may earn us a commission at no extra cost to you, and by using this site, you agree to our terms and privacy policy.

What is Google Gemini?

Google Gemini is a new tool from Google DeepMind that’s all about making artificial intelligence (AI) smarter and more useful. Unlike other AI systems that just work with text, Gemini can understand and work with different types of information like pictures, sounds, and even computer code, all from one place.

From PaLM 2 to Gemini

Before Gemini, Google had another AI system called PaLM 2, which was part of Google Bard. In December 2023, Google introduced Gemini, which took over from PaLM 2. Then, in February 2024, Google announced that Bard would now be called Gemini. This change shows how important Gemini is to Google’s plans for AI.

Why Gemini is special

The cool thing about Gemini is that it can handle lots of different types of information, not just words. This means it can help with more things, like understanding pictures or helping write computer code. By putting all these abilities into one system, Gemini makes using AI more straightforward and powerful.

What Gemini means for us

Google Gemini is a big step forward in making AI more helpful in everyday life. It could change how we search for information, use digital assistants, or create new software. As Gemini gets better, we’ll start to see AI doing more things more smartly, making our interaction with technology easier and more natural.

The Google Gemini ecosystem: A new era of AI integration

Sundar Pichai, the CEO of Google and Alphabet, shared some exciting news about Gemini. He said that Gemini is not just another tool; it’s going to be a big deal for everyone. Gemini will be part of everything, from the Google products that millions of us use every day to special tools for developers and businesses to make cool new stuff.

A new name for familiar tools

Google is shaking things up by changing the names and combining many of its AI tools under the Gemini name. For example, Duet AI will now be known as Gemini for Workspace. Plus, Gemini Advanced is getting mixed in with other tools we use all the time, like Gmail, Docs, Sheets, Slides, and Meet.

This change means that Google’s smart AI technology will be part of the tools many of us use for work and school. Whether you’re writing an email, working on a spreadsheet, or having a video meeting, Gemini’s AI will be there to help make things easier and smarter.

How Gemini brings AI to life

Gemini AI, the smart technology behind Google’s latest innovation, is believed to be built on something called the Google Pathways architecture. A group of separate machine learning (ML) models are first trained to carry out a particular job. Imagine this as building blocks of smart learning that, once they learn how to do something well, can be connected together like pieces of a puzzle.

In this setup, different blocks or modules are first taught to master a specific task. After they’ve learned their part, these modules are linked to form a bigger network. This network is pretty flexible; the modules can either work solo or team up to tackle various tasks, creating different types of results.

How Gemini understands and responds

On the technical side, there are tools called encoders that help translate different types of data (like pictures, text, or sounds) into a language that all parts of the AI can understand. Then, decoders come into play, taking this common language and transforming it into the final output, whether that’s writing, images, or something else, depending on what you need.

It’s also noted that these AI models work faster and smoother on Google’s own super-powered computers known as Tensor Processing Units (TPUs).

Making AI user-friendly

Despite all the complex stuff happening behind the scenes, Gemini is designed to be easy for anyone to use. It has a simple interface that lets people use these generative AI models without needing to be tech experts. Whether you’re creating something new or just exploring what AI can do, Gemini is made to be accessible for all kinds of users.

Gemini AI models

Zoubin Ghahramani, a big name at Google DeepMind, once thought that Gemini would come in four sizes, just like PaLM 2 did: Gecko, Otter, Bison, and Unicorn. Each of these models had its own special use:

  • Gecko was going to be super lightweight, perfect for using on your phone.
  • Otter was aimed at doing a bunch of different tasks that only need one type of data (like just text or just images).
  • Bison was set up for doing a few tasks that mix different types of data (like text and images together).
  • Unicorn was the big deal, designed to handle lots of complex tasks mixing different types of data.

A new approach: Three sizes of Gemini

But plans change, and now it looks like Gemini is going to be simpler, available in just three versions:

  • Gemini Nano is made for mobile. It’s like Gecko, meant to be easy to use on your phone without eating up all your battery.
  • Gemini Pro is for your computer or browser. It’s the go-to choice for a broad range of tasks, whether you’re working, studying, or just playing around.
  • Gemini Advanced (sometimes called Gemini Ultra) is the premium option for people who subscribe. It’s packed with features for serious tasks, perfect for developers, creators, and anyone who needs the extra power.

This streamlined lineup makes it easier to choose the right version of Gemini for what you need, whether you’re on the go with your phone, at your desk, or looking for advanced AI capabilities.

Training the brains behind Gemini AI

How Gemini AI is trained

The training of Gemini’s large language models (LLMs) is a bit of a tech mystery, combining several advanced techniques:

  • Supervised Learning : Here, Gemini AI was like a student learning from a book with answers in the back. It practiced predicting outcomes based on examples where the right answers were already marked.
  • Unsupervised Learning : This time, Gemini AI was more like a detective, looking for clues and patterns all by itself, without any clear answers given. It had to make sense of data without any guidance on what was right or wrong.
  • Reinforcement Learning : Think of this as learning to ride a bike with trial and error, getting better over time. Gemini AI tried different things, learned from mistakes, and figured out how to achieve its goals, earning rewards and avoiding mistakes.
  • The Secret Sauce : Reinforcement Learning with Human Feedback

Some experts think Google really leaned into a special technique called reinforcement learning with human feedback (RLHF) while training Gemini, using powerful Cloud TPU v5e chips. Google says these chips are way more powerful than the ones used for training something like ChatGPT, making Gemini AI super smart.

What Gemini AI learned from

While Google’s been pretty quiet about exactly what data they used to teach Gemini AI, it’s believed they might have used the LangChain framework and reused some of the same data from training their earlier model, PaLM 2.

This means Gemini could have started learning from a mix of stuff from the internet, books, computer code, pictures, sounds, and videos. The big question is whether teaching Gemini AI with all this mixed data from the start works better than focusing on one type of data at a time and then adding more.

Either way, Gemini AI is built to handle different types of information right from the get-go, showing just how advanced and flexible it is.

The story behind Google Gemini’s name

There’s been a lot of guesses about why Google chose the name “Gemini” for its latest AI project. Some thought Gemini stood for “Generalized Multimodal Intelligence Network Interface,” but no one’s been able to prove that’s true.

A nod to the stars and myths

A more likely explanation comes from Google Bard, suggesting that the name “Gemini” was inspired by the stars – specifically, the Gemini constellation and the Greek myth of Castor and Pollux that it represents. This wouldn’t be a surprise since Google has a knack for picking names related to space for its products.

When asked, Google Gemini itself hinted that the name does indeed come from the constellation and myth, fitting right in with Google’s tradition of space-themed names for its projects. This choice reflects not just a creative naming pattern but also hints at the power and connectivity the Gemini project aims to embody, much like the twin brothers Castor and Pollux, known for their teamwork and unity.

How much does Gemini AI cost?

Free access to Gemini

Whether you’re using a computer or a phone, you can start using Gemini for free. If you’re on a desktop, just hop onto your browser, and you’re good to go. For folks with Android phones, there’s a Gemini app waiting for you. And if you’re an iPhone user, you can get into Gemini through the Google app.

Stepping up with Google One AI premium

For those who want more out of Gemini, there’s a premium version you can subscribe to, called Google One AI Premium. This isn’t just any subscription – it’s a special level of the Google One service, which many people already use for extra storage space.

Signing up for Google One AI Premium to get Gemini Advanced will cost you $19.99 a month. But there’s a sweet deal to start: a two-month free trial to see if you like it. Plus, if you decide to go for the whole year, there’s a discount waiting for you.

So, whether you’re just dipping your toes into what Gemini can do or you’re ready to dive deep with the advanced features, there’s an option that fits your needs and budget.

Related terms

Related articles

About XPS's Editorial Process

XPS's editorial policy focuses on providing content that is meticulously researched, precise, and impartial. We adhere to rigorous sourcing guidelines, and every page is subject to an exhaustive review by our team of leading technology specialists and experienced editors. This method guarantees the integrity, pertinence, and utility of our content for our audience.

Ross Jukes
Ross Jukes
Ross Jukes is an accomplished American copywriter with a Bachelor’s Degree in English Literature and a minor in Creative Writing. Based in the United States, Ross is a language expert, fluent in English and specializes in creating compelling and engaging content. With years of experience in the industry, he has honed his skills in various forms of writing, including advertising, marketing, and web content. Ross's creativity and keen eye for detail have made him a valuable asset in the field of copywriting, where he continues to excel and innovate.

Why Trust Us

Our editorial policy emphasizes accuracy, relevance, and impartiality, with content crafted by experts and rigorously reviewed by seasoned editors for top-notch reporting and publishing standards.

Purchases via our affiliate links may earn us a commission at no extra cost to you, and by using this site, you agree to our terms and privacy policy.

Popular terms

What is HRIS?

HRIS, short for Human Resource Information System, is a software platform that allows employers to store and manage employee data in an easily accessible...

What is Market Capitalization?

Market capitalization or market cap is a financial measure that denotes the value of a digital currency. It has historically been used to measure...

What is a WebSocket

In the world of web development, communicating between clients and servers in real time has become a necessity. That's where WebSocket comes in, using...

What is AI Ethics?

AI ethics is a field that is concerned with the creation and employment of artificial intelligence (AI). It is a set of values meant...

What is Relative Strength Index (RSI)?

Relative Strength Index (RSI) is a powerful technical analysis tool which is used as a momentum oscillator for measuring how fast and how much...

Latest articles