A week in Generative AI: Gemini 3, Group Chats & Nano Banana
News for the week ending 23rd November 2025
For the first time this year, I think it’s safe to say that the week belongs to Google with the release of Gemini 3 and Nano Banana Pro. Not to be outdone, OpenAI also decided to roll out ChatGPT Group Chat to everyone globally after just three days of testing.
In Web 4.0 news there’s a great article from The Verge on how AI browsers are a huge threat to Amazon, Google has a blog post on AI shopping for the holidays, and Amazon blocks more of OpenAI’s web crawlers in a clear sign that they want to be the ‘closed ecosystem’ vs. OpenAI’s ‘open ecosystem’ in the coming AI commerce battle.
On the Ethics front there were reports that Europe is planing to scale back it’s privacy and AI laws, that Google has been using Gmail to train its AI models, and UK consumers were warned over taking financial advice from AI chatbots.
There’s a good Long Read from Wired with an interview with Fidji Simo, the CEO of Applications at OpenAI and an interesting article on how electricity might be the new currency in the future.
Google DeepMind launches Gemini 3 with new coding app and record benchmark scores
This is a big launch for Google DeepMind, who have been relatively quiet this year whilst OpenAI have been shipping faster than any technology company that I’ve ever seen, which isn’t always a good thing!
Gemini 3 is being positioned as a new series of models, with lots more to be announced in the coming weeks and months. For now Google is going big on intelligence/reasoning, coding, and generative interfaces.
Intelligence & Reasoning
Google claim they’re seeing a big jump in reasoning with Gemini 3 Pro vs. Gemini 2.5 Pro and this is backed up by the impressive benchmark scores they’ve shared for their new model. For example, Gemini 3 Pro has beaten the previous state of the art model, GPT-5 Pro, on Humanity’s Last Exam by a comfortable 18%.
Coding
Alongside the new model, Google has also released something called Antigravity, which is Google’s answer to OpenAI’s Codex and Anthropic’s Claude Code. It’s an agentic software development platform that enables coding agents to autonomously plan and execute complex end-to-end coding tasks, sound familiar? The twist with Antigravity is that it has three panes - one for prompting, one for coding, and a browser window that can show the code in action. Sounds like a great twist and I’m looking forward to trying it out.
Generative Interfaces
This is the biggest one for me and I do think that this is the future of the incredibly basic ‘chat’ interface we have grown accustom to over the last 3 years. Gemini 3 can create “generative interfaces” on the fly that give users fully interactive experiences, not just text. These are rolling out in a few places to start off with:
The Gemini app where the model designs and codes a custom interactive response per prompt
Google Search’s AI mode where the model can build interactive visual explanations and tools
We’ve got a way to go before these generative interfaces become the norm, but I do think its the future of not only how we interact with AI platforms, but also how we interact with digital services and content in general.
If you want to read a bit more about the release of Gemini 3, below are links to two of my favourite commentators with their takes:
One Useful Thing - Three Years from GPT-3 to Gemini 3
Simon Willison - Trying out Gemini 3 Pro with audio transcription and a new pelican benchmark
ChatGPT launches group chats globally
Well that didn’t take long…. they only started testing them about three days before they announced they’re rolling them out globally!
As I said last week, I think this is another example of OpenAI wanted to rebuild the entire internet on top of ChatGPT and they’re steadily ticking off major use cases. This ticks off messaging, which according to Ofcom is something 94% of UK adults do on the daily. Be interesting if OpenAI can get traction with this over the coming months. I think they’ll probably need a standalone app for it to really take off, but let’s see…
Google releases Nano Banana Pro, its latest image-generation model
Just a couple of months after the release of the original Nano Banana image model, Google just dropped the Pro version that has new editing capabilities, is able to generate higher resolutions, has better text rendering, and ability to search the web for inspiration.
Nano Banana Pro is built on Gemini 3, so is similar to ChatGPT’s image generation capabilities in that it can more accurately interpret and follow prompts as well as reason over instructions given to it which powers its new editing capabilities.
As with most image generating models now, safe guards and copyright issues still abound. The Verge has a good article on some of the challenges with Nano Banana Pro, but they’re no different to similar issues faced by all image generators and OpenAI’s recently upgraded Sora 2 video model. Because the genie already seems to be out of the bottle with this technology, I honestly think the only answer is that people will become more attuned to generated content and be able to spot it more easily. We still need some updated and more nuanced copyright laws though.
There is also some good coverage and examples from Simon Willison - Nano Banana Pro aka gemini-3-pro-image-preview is the best available image generation model
Web 4.0
The DoorDash Problem: How AI browsers are a huge threat to Amazon
Amazon quietly blocks more of OpenAI’s ChatGPT web crawlers from accessing its site
Google’s AI Mode can now help you visualize your travel plans
Intuit signs $100M+ deal with OpenAI to bring its apps to ChatGPT
AI Ethics News
Google denies ‘misleading’ reports of Gmail using your emails to train AI
AI firms must be clear on risks or repeat tobacco’s mistakes, says Anthropic chief
Becoming an AI detective is a job I never wanted and wish I could quit
UK consumers warned over AI chatbots giving inaccurate financial advice
Klarna says AI drive has helped halve staff numbers and boost pay
Paul McCartney joins music industry protest against AI with silent track
Warner Music settles copyright lawsuit with Udio, signs deal for AI music platform
AI is changing the relationship between journalist and audience. There is much at stake
You Must Read This Riveting Whistleblower Lawsuit About Allegedly Dangerous Robots
Long Reads
Wired - OpenAI’s Fidji Simo Plans to Make ChatGPT Way More Useful—and Have You Pay For It
Electrek - Electricity is about to become the new base currency and China figured it out
The Guardian - What AI doesn’t know: we could be creating a global ‘knowledge collapse’
“The future is already here, it’s just not evenly distributed.“
William Gibson