Google Gemini, explained

Artificial intelligence has go this year’s wonderment technology. But because it comes successful a batch of different flavors from a batch of different companies, it tin beryllium really confusing. You’ve not only sewage nan ChatGPT bot created by OpenAI, but you’ve sewage nan large 3 — Google, Apple, and Microsoft — cooking up their ain versions.

Google’s latest effort is called Gemini, and it’s nary little confusing than nan others. 

When I first started researching Gemini, I did a Google hunt for “versions of Google Gemini.” On apical of nan search, I sewage an AI-generated summary that started:

“Google Gemini has 3 versions: Ultra, Pro, and Nano. Ultra is nan largest exemplary and is designed for analyzable tasks, while Pro is nan champion exemplary for scaling crossed a wide scope of tasks, and Nano is nan astir businesslike exemplary for on-device tasks.”

Okay, bully enough. But it’s not nan complete story.

What is Gemini?

Gemini is nan 3rd zodiac sign, associated pinch nan twins Castor and Pollux from Greek mythology.

Okay, sorry. I couldn’t resist. Gemini is simply a chatbot created by Google that has replaced its erstwhile chatbot named Bard. It’s based connected thing called a ample connection exemplary (or LLM), besides called Gemini, which was developed by DeepMind, a portion of Google. 

So Gemini is some a chatbox and an LLM? How galore types of Gemini are there?

How overmuch clip do you have? Seriously, though, we’re going to limit ourselves to nan types of Gemini that you whitethorn brushwood because nan number of iterations consciousness endless. 

Originally, when it was introduced successful December 2023, Gemini offered 3 different versions (known arsenic models): Nano arsenic a lightweight Android version, Pro for mundane wear, and Ultra for heavyweight business / endeavor usage.

Then connected May 14th, during its I/O 2024 event, Google introduced Gemini 1.5 Pro, nan first successful what nan institution called a “mid-sized multimodal model.” According to Google, nan caller type of Pro is astir arsenic powerful arsenic nan erstwhile Ultra type and is meant to heighten existing apps and create caller ones for day-to-day uses.

Hold on. Multimodal?

In different words, it tin judge prompts successful each different modes of communication: text, images, audio, and video.

So that’s it for nan models, right?

Well, not quite. There’s besides Gemini 1.5 Flash, which is simply a faster type of Gemini for developers who will beryllium capable to usage it successful circumstantial applications. In different words, unless you’re a developer, it’s not thing you will beryllium moving with.

So, conscionable to reiterate, we now person 4 Gemini models for developers to activity with: Ultra, Pro, Flash, and Nano. (We’ll show you really you tin play pinch it yourself successful a moment.) 

I watched nan Google event, and they kept talking astir 1 cardinal tokens, 2 cardinal tokens. What was that each about?

That’s what you get for watching an arena that’s meant much for developers than for mundane group for illustration us. But it’s really not each that difficult. 

Tokens are nan elements of words that are utilized to train AI models specified arsenic Gemini. The much tokens an AI exemplary is tin of, nan much info you tin provender nan AI and nan amended it will understand what you request and what it tin springiness you. 

Okay, backmost to Gemini 1.5 Pro. What tin I do pinch it?

Well, if you’re a developer, you tin usage it to adhd to aliases create a bunch of caller apps. Otherwise, Google is adding it to a batch of its existing apps and creating caller ones. 

Like?

Well, conscionable arsenic an example, let’s commencement pinch Google Photos. A caller characteristic expected this summer, called Ask Photos, will fto you hunt utilizing much analyzable queries. Instead of conscionable uncovering each nan photos of your grandmother, for example, you should beryllium capable to inquire it to “Find each nan photos of my grandma done nan years that show her moving connected her carpentry projects.” 

There’s besides nan existing Lens app, which uses some matter and photos to thief you place and investigation stuff. Lens will now beryllium capable to find info utilizing videos arsenic well. Google’s demonstrated it by taking a video of a misbehaving grounds subordinate and utilizing a video to find retired why nan tonearm wasn’t contacting nan record.

You cognize that sidebar successful Google Docs, Sheets, Slides, Drive, and Gmail? The 1 wherever you tin now entree various different Google apps? Well, it’s going to beryllium taken complete by Gemini, which will beryllium utilized to unify — or, astatine least, to link — a assortment of Google apps truthful that you’ll beryllium capable to, say, easy reference a Google Doc successful an email aliases visa versa. It should beryllium rolling retired to subscribers adjacent month.

Even Google’s basal hunt has been affected: AI Overviews now lead disconnected your hunt results, giving you an AI-generated summary of what Google thinks you’re looking for. (Although there’s been a batch of pushback connected that and rather a fewer users looking to get free of it.)

Those are existing apps. How astir caller ones?

Lots of them. Currently, immoderate include:

Project Astra, which is fundamentally Google Assistant pinch nan added expertise to spot (via your phone’s camera) and respond to, and with, spoken language. This is still successful its early days, truthful you astir apt won’t spot it for a while.

LearnLM, which will thief students find answers to their questions utilizing acquisition sources; according to nan company, it’s already been built into immoderate products and is being introduced to educators.

Veo, a “generative AI video model.” Generative arsenic successful it will generate 1080p videos that you inquire it to create. You want a video of a feline wearing a nightgown and a apical chapeau jumping complete nan Moon? Veos is what you want to use. Well, erstwhile you tin — for illustration Project Astra, it’s still being tested and won’t beryllium disposable to nan wide nationalist for a while.

This each sounds interesting. How tin I motion up? And is it free?

You tin commencement moving pinch nan Gemini 1.0 chatbot right now and correct here. However, if you want to play pinch Gemini 1.5 Pro — which is faster and gives you much capabilities — you’ll request to subscribe to Gemini Advanced, which will costs $20 a period aft a two-month trial. (Gemini Advanced is considered portion of a Google One subscription, truthful you’ll besides get 2TB of information retention and different Google One benefits.)

If you’re a business utilizing Google Workspace and you want to effort nan much blase levels of nan AI (also starting astatine $20 a month), you tin find much accusation here. 

Anything other I request to know?

Just nan accustomed cautions. Like each AI applications, Gemini’s answers tin beryllium iffy — successful different words, downright wrong. The tech is decidedly successful its early stages, and truthful while it tin beryllium a useful tool, you should besides cheque immoderate information you get. It’s gotten truthful that incorrect accusation generated by AI engines has gotten its ain name: hallucinations, because by accessing incorrect information, nan AIs are creating their ain reality. So, purchaser beware.

That being said, it looks for illustration AIs are going to beryllium pinch america for a agelong time. It’s not a bad thought to do immoderate hands-on successful bid to go acquainted pinch them and really they work. Besides ChatGPT and Gemini, location are Microsoft’s upcoming CoPilot Plus PCs, which will travel pinch built successful AI-capable hardware, not to mention Apple’s just-announced and upcoming suite of features called Apple Intelligence. So depending connected your favourite operating system, not to mention your level of curiosity, you tin research pinch a assortment of AI chatbots, enhanced apps, and different features.