Google Gemini, explained – The Verge

Artificial intelligence has become this year’s wonder technology. But because it comes in many different flavors, from many different companies, it can be really confused. You not only have the ChatGPT bot created by OpenAI, but also the big three – Google, Apple and Microsoft – creating their own versions.

Google’s latest attempt is called Gemini and it’s no less confusing than the others.

When I first started researching Gemini, I did a Google search for “Google Gemini versions”. At the top of the search, I received an AI-generated summary that began:

“Google Gemini has three versions: Ultra, Pro and Nano. Ultra is the largest model and is designed for complex tasks, while Pro is the best model for scaling a wide range of tasks and Nano is the most efficient model for on-device tasks.”

Okay, good enough. But it’s not the complete story.

Gemini is the third sign of the zodiac, associated with the twins Castor and Pollux from Greek mythology.

OK sorry. I could not resist. Gemini is a chatbot created by Google that replaced its previous chatbot called Bard. It’s based on something called a large language model (or LLM), also called Gemini, which was developed by DeepMind, a part of Google.

How much time do you have? Seriously, let’s limit ourselves to the types of Gemini you can find because the number of iterations seems infinite.

Originally, when it launched in December 2023, Gemini offered three different versions (known as models): Nano as a lightweight Android version, Pro for everyday use, and Ultra for heavy commercial/enterprise use.

Then, on May 14, during its I/O 2024 event, Google introduced Gemini 1.5 Pro, the first in what the company called a “midrange multimodal model.” According to Google, the new version of Pro is as powerful as the previous version of Ultra and aims to enhance existing apps and create new ones for everyday use.

In other words, it can accept notices in all different modes of communication: text, images, audio and video.

Well, not exactly. There is also Gemini 1.5 Flash, which is a faster version of Gemini for developers who will be able to use it in specific applications. In other words, unless you’re a developer, it’s not something you’ll work with.

So, just to reiterate, we now have four Gemini models for developers to work with: Ultra, Pro, Flash, and Nano. (We’ll tell you how you can play with this in a moment.)

That’s what you get for attending an event that means more to developers than regular people like us. But it’s really not that difficult.

Tokens are the word elements used to train AI models like Gemini. The more tokens an AI model is capable of, the more information you can feed the AI ​​and the better it will understand what you need and what it can offer you.

Well, if you are a developer, you can use it to add or create various new applications. Otherwise, Google will add it to many of its existing apps and create new ones.

Well, just as an example, let’s start with Google Photos. A new feature expected this summer called Ask Photos will let you search using more complex queries. Instead of just finding all the photos of your grandmother, for example, you should be able to ask to “Find all the photos of my grandmother over the years that show her working on her woodworking projects.”

There’s also the existing Lens app, which uses text and photos to help you identify and search for things. Lens will now also be able to find information using videos. Google demonstrated this by recording a video of a misbehaving record player and using the video to figure out why the tonearm wasn’t in contact with the record.

You know that sidebar in Docs, Sheets, Slides, Drive, and Gmail? The one where you can now access tons of other Google apps? Well, it will be taken over by Gemini, which will be used to unify – or at least connect – a variety of Google apps so that you can, say, easily reference a Google Doc in an email. or vice versa. It should be released to subscribers next month.

Even basic Google search has been affected: AI overviews now open search results, providing an AI-generated summary of what Google thinks you’re looking for. (Although there has been a lot of resistance to this and some users wanting to get rid of it.)

A lot of them. Currently, some include:

Astra Project, which is essentially Google Assistant with the added ability to see (via your phone’s camera) and respond, and with spoken language. This is still in its early days, so you probably won’t see it for a while.

LearnLM, which will help students find answers to their questions using educational sources; according to the company, it is already integrated into some products and is being presented to educators.

Veo, an “AI generative video model”. Generative as it will be generate 1080p videos you asked to create. Want a video of a cat in a sweater and top hat jumping on the moon? Veos is what you want to use. Well, when you can – like Project Astra, it’s still being tested and won’t be available to the general public for a while.

You can start working with the Gemini 1.0 chatbot now and here. However, if you want to play with Gemini 1.5 Pro – which is faster and offers more features – you will need to sign up for Advanced Gemini, which will cost $20 per month after a two-month trial. (Gemini Advanced is considered part of a Google One subscription, so you’ll also get 2TB of data storage and other Google One benefits.)

If you’re a company using Google Workspace and want to try the most sophisticated levels of AI (also starting at $20 per month), you can find more information here.

Just the usual care. Like all AI applications, Gemini’s answers can be unreliable – in other words, flat out wrong. The technology is definitely in its early stages, and while it can be a useful tool, you should also verify any data you obtain. It has gotten to the point that the misinformation generated by AI engines has gained its own name: hallucinations, because by accessing misinformation, AIs are creating their own reality. So buyer beware.

That said, it looks like AIs will be with us for a long time. It’s not a bad idea to practice a little to familiarize yourself with them and how they work. In addition to ChatGPT and Gemini, there are Microsoft’s upcoming CoPilot Plus PCs, which will come with built-in AI-capable hardware, not to mention Apple’s recently announced and upcoming suite of features called Apple Intelligence. So, depending on your favorite operating system, not to mention your level of curiosity, you can try out a variety of AI chatbots, enhanced apps, and other features.