Google DeepMind Introduces Gemini: A Fun New AI to Rival GPT-4
LONDON — Google DeepMind rolled out Gemini, a multimodal large language model, on December 6, 2023, aiming to top OpenAI’s GPT-4.
Launched from DeepMind’s base in London, this AI can handle text, images, audio, and more. It’s built to compete in the fast-moving world of smart machines.
A Big Leap for AI Fans
Gemini isn’t just another chatbot. It’s a family of models — Ultra, Pro, and Nano — each made for different jobs. Google says Ultra beats GPT-4 in 30 out of 32 tests, like math and coding. The launch came fast after months of buzz, showing Google’s push to lead AI again. It’s already in Bard, Google’s chat app, as of March 13, 2025.
The rollout has three flavors. Nano runs on phones, Pro powers Bard, and Ultra is for tough tasks. Developers get Pro now, while Ultra hits early 2024. Why multimodal? It can see, hear, and talk — not just type. This makes it handy for real-life stuff, like explaining a photo or coding an app.
How Gemini Stacks Up
Google’s been playing catch-up. OpenAI’s ChatGPT, built on GPT-4, stole the spotlight in 2022. Now, Gemini steps in with fresh moves. It’s trained on Google’s own chips, called TPUs, not Nvidia’s gear. That cuts costs and boosts speed. Plus, it’s not stuck on text — it mixes images and sound too.
The Backstory of the Battle
This isn’t Google’s first shot. Its older model, PaLM 2, tried to match GPT-4 but fell short. DeepMind, bought by Google in 2014, teamed up with Google Brain to make Gemini. The name nods to NASA’s twin-astronaut missions and the Brain-DeepMind merger. “It’s our most capable model yet,” said Demis Hassabis, DeepMind’s CEO, in a blog post.
What It Can Do
Gemini shines in demos. It solves homework, writes code, and chats about videos — all at once. Ultra scored 90% on a big test called MMLU, topping human experts at 89.8%. GPT-4 got 87%. Coding’s a win too — Gemini’s AlphaCode 2 beats 85% of human coders, up from 50% last year. But some say the demos look too polished.
Critics aren’t all in. Tech watchers like MIT’s review team say Gemini’s edge over GPT-4 is slim. Multimodal tricks aren’t new — GPT-4 added them too. And Google’s video demo? Not live, admits the company. Still, the buzz is real — Google’s stock popped 5.3% after the news.
Why It’s a Big Deal
The AI race is hot. OpenAI, backed by Microsoft, set the pace. Google’s fighting back with Gemini. It’s not just about bragging rights — this tech could change how we search, work, and play. Think smarter phones, better apps, even science breakthroughs. DeepMind’s past wins, like beating Go champs, fuel the hype.
But there’s a catch. Gemini’s not wide open yet. Safety checks come first, says Google. Only Bard uses Pro now, and Ultra’s locked for testers. The EU and UK miss out due to rules. That’s left some fans grumbling — they want it now. Meanwhile, rivals like Anthropic and Meta watch closely.
What’s Next for Gemini?
Google’s got plans. Gemini will hit Search, Ads, and Chrome soon. A souped-up Bard Advanced with Ultra is due in 2024. Developers can tinker with Pro via APIs already. If it keeps up, Gemini might shift AI’s power balance. OpenAI’s likely cooking GPT-5, though — the fight’s far from over.
This could spark a tech boom. More multimodal AIs might pop up, mixing sight and sound for all. Or it could flop if bugs or rules slow it down. For now, Gemini’s a shiny toy with big dreams. Will it rule AI? Time — and users — will tell.