Today on AI For Humans:
Google’s Turn For Next Gen AI
Disney Signals AI Content Coming
Plus, Agentic AI For Everyday Use!
Love this newsletter? Forward it to one curious friend. They can join in one click.
Welcome back to the AI For Humans newsletter!
It’s coming. The hype is real. But will Gemini 3 Pro be all that?
In the internet’s worse kept secret, Google is prepping to release the next generation of both their Gemini and Nano Banana AI models this week and maybe more. Rumors have both looking to be a massive upgrade from the current SOTA.
How sure are we? Well, Google CEO Sundar Pachai is tweeting Polymarket charts…
Over the past few days, Google employees have taken to vague-posting ‘get ready’ sorts of messages about the release.
It’s funny, these posts are so vague that I can’t find them now on X but I swear they’ve filled my feed. AI For Humans’ favorite OpenAI dev Roon has seen them too.
Supposedly, Gemini Pro 3.0 has already maybe purposefully been leaked inside the Canvas function of the Gemini iOS app & website. I spent some time over the weekend prompting a new vaporwave-inspired website design that def seems better than what was possible with Gemini 2.5 Pro.
The betting money (for now) is on a Tuesday release and we’re excited. We might have to do a second newsletter this week just for that but something else Google showed last week is a much bigger deal than it’s getting credit for…
Sima 2: AI Agents Operating in Complex Worlds
Maybe to get ahead of the Gemini 3 launch, Google dropped a couple pretty big things that are more impressive than they might look at first glace.
First, they updated their real-time AI audio model in Google Live with new voices & better response times. It’s cool & live in the Gemini app right now.
But WAY more interesting is Sima 2.
In the simplest explanation, Sima 2 is an AI agent that plays video games.
You might wonder: So what? I play video games. Big deal.
Well, when you start digging in on what Sima 2 is really doing, it points to a new paradigm for AI and a potential path to AGI.
First and foremost, Sima 2 plays video games entirely on its own. It gets dropped into complicated games like No Man’s Sky or Minecraft and uses Gemini’s reasoning ability to figure out what to do (while still chatting with a human companion).
Human input can tell the Sima 2 system what to do… and it does it.
However, the much bigger deal is that Sima 2 is showing signs of self-improvement in how it operates within these games. The longer it plays and the more of the enviroment and rules it learns, the better it gets at operating withing them.
There are some echoes here to Google Deepmind’s work on Atari Consoles way back in 2020 which showed that specific AI could learn to get better at these systems over time.
But these current game environments are MUCH larger and more complex and are, interestingly, not that far off from the real world.
Simulated Worlds + Intelligent Agents = AGI Training
The coming upgrade in Gemini is an important part of what Sima 2 is able to do and how it will get better.
As AI reasoning agents improve, they start to make better decisions thinking through their actions. Speed makes a big difference here (we don’t take multiple minutes to decide to do something) but that’s improving quickly as well.
I suggest you read through the Sima 2 blog post, especially the end, where the DeepMind team discusses how this system will help improve AI’s embodied intelligence.
Thank you Google AI overviews…
In short, Gemini 3 Pro & these weird video game models might just be training your home robot of the future.
Hopefully, this will help them get better at shutting the dishwasher in a few years.
See you next week!
- Gavin
In this week’s AI For Humans: OpenAI dropped GPT-5.1 which is…good?👇
3 Things To Know About AI Today
Disney CEO Bob Iger Says AI UGC Is Coming
In this week’s corporate earnings call, Bob Iger announced that that Disney+ was “in the midst of rolling out the biggest and the most significant changes - from a product perspective, from a technology perspective - since we launched the service in 2019.”
And, in 2025, that means an AI content strategy.
If you’ve been following us for a while, you knew this was coming but a lot of people still seem very unhappy that Disney is leaning into AI content.
It’s pretty obvious that Disney won’t pursue this entirely on their own. They don’t have the AI video model to power it and the value to them is in the IP expansion (and maybe a new licensing fee structure).
Will they partner with OpenAI’s Sora? Or… could the big Google announcements this week introduce a Veo-powered something? We shall see.
AI Science Is (Maybe) Here?
We’ve been waiting for the AI news-cycle to shift from ‘Ugh, AI video sucks’ to ‘Hey AI might be good for humanity!’ and it might finally be happening.
Edison Scientific, an AI science start-up, announced their next generation AI scientist model Kosmos and it got Sam Altman pretty excited.
Kosmos isn’t just spitting out vibes. In early tests, it tore through more than a thousand papers, doing six months of human research in one go.
Our beta users estimate that Kosmos can do in one day what would take them 6 months, and we find that 79.4% of its conclusions are accurate.
Even wilder, it landed seven legit discoveries, four completely new to science, which gives us AI fans quite a bit of hope for a better narrative.
AI For Humans New AI Word Alert: AI Slop Vs AI Kino
Here’s a phrase you might not know: AI Kino.
This version of the term kino has its origins in internet film culture as a sort of insult towards film snobs that has morphed into a compliment of high culture content masquerading as low culture.
If nothing else, the video above should elucidate a feeling in you that’s remarkably different than your average AI slop post.
And I think that is AI kino.
We 💛 This: Agentic AI In Everyday Life
Most non-AI native people understand how to use AI tools like ChatGPT in the same way they’ve used Google but I’m here to tell you that even today these tools can do a lot more for your everyday life via their agentic capabilities.
My wife and I are packing up our home right now (We’re moving! And we’re not entirely sure where yet!) and we’ve been Marie Kondo-ing our lives by selling a bunch of our old stuff we either don’t use or don’t want anymore.
One of the things on my list was to look at off-loading a handful of my old comics that I’ve been moving from place to place and I just handed it off to GPT-5.1.
This worked incredibly well and let me know that one of these comics (Incredible Hulk #340) is prob worth quite a bit but more importantly it saved me a ton of time.
I’ve also been using it to write Facebook Marketplace posts. I upload photos & dimensions, ask it to find the value, write the post etc. It just works.
The future we were promised is starting to happen and that’s exciting.
Are you a creative or brand looking to go deeper with AI?
Join our community of collaborative creators on the AI4H Discord
Get exclusive access to all things AI4H on our Patreon
If you’re an org, consider booking Kevin & Gavin for your next event!