😺Microsoft's new AI games...

PLUS: Microsoft built a new quantum chip?!

Welcome, humans.

Had to share this one, because c’mon—it’s cats.

In all seriousness, video AI is getting preeetty convincing.

ByteDance just released a new paper for a model called Phantom that can generate video with consistent actors and products (here’s 10 demos).

There’s also Pika 2.1, which is ABSOLUTELY wild, and has a new feature called Pikadditions that does a similar thing. In the (paraphrased) words of Roaring Kitty: “we like the tool.”

Here’s what you need to know about AI today:

Microsoft built an AI that generates legit gameplay videos.
Claude beat OpenAI models in developer coding benchmark.
Apple launched budget AI phone for $599.
Microsoft also built a new quantum chip.

Advertise in The Neuron here.

Microsoft just dropped a gaming AI that can “dream up” new gameplay videos.

Large language models (chatbots) are great, but it seems like we’re now in a waiting period before anything really new comes out.

Yeah, there’s Grok 3, but unless you’re an Elon-stan, that doesn’t mean much. AI race heating up, competitors gaining on OpenAI, etc, etc. Been there, wrote that.

Let’s talk about what’s going on in AI OUTSIDE of language models…

And on that front, there’s some actually cool stuff going on. Exhibit A: Yesterday, Microsoft Research revealed Muse, their first AI model designed specifically for generating gameplay videos (paper).

Muse is based on something Microsoft calls a “World and Human Action Model” (WHAM!). Think of it like GPT, but instead of predicting the next word in a sentence, it predicts how a game world will respond to player inputs.

Microsoft trained Muse on 7+ years of continuous gameplay data from a single game—Bleeding Edge, a 4v4 team brawler from Xbox's Ninja Theory studio. That's over a billion frames of gameplay and controller inputs, FYI.

Here's what makes Muse special:

It takes a single game frame and generates multiple different ways the gameplay could continue.
It understands game physics and rules (characters don't walk through walls; gravity works properly).
You can edit the scene (like adding new characters) and Muse will maintain those changes in the generated sequence.
You can even use a controller to guide where you want the character to go.

The demos are pretty cool. In one example, the researchers took a promotional image from Bleeding Edge, added an extra character that wasn't there originally, and Muse generated a completely new sequence naturally incorporating that added character.

Why this matters: Game development is expensive and time-consuming. A tool that can quickly generate different gameplay scenarios could be huge for:

Game designers testing new ideas.
Developers creating training data for “world model” AI.
QA teams exploring edge cases.
Artists visualizing different character animations.

What's next: Microsoft is open-sourcing the model weights and demo app on Azure AI Foundry. This means other researchers and developers can start experimenting with it right away. While it currently only works with one game, imagine future versions that could…

Preserve classic games by learning their rules and recreating them for modern hardware.
Generate NEW content for existing games or help designers prototype ideas.
Create endless variations of game levels for max replay value.

Our take: The interesting thing here is the hard problem of generating interactive content that follows consistent rules. And the fact that Microsoft is open-sourcing it suggests they're more interested in advancing the field than keeping the tech proprietary (are you shocked pikachu? Same).

The big question remains: Will Muse grow beyond a research demo into something game developers actually use?

Game studios are notoriously cautious about new tech, but with Microsoft's Xbox division already exploring ways to use it, we might see practical applications sooner than you think.

However, that might not be good for an already damaged video game industry…

FROM OUR PARTNERS

Here’s one AI tool we use every week at The Neuron.

Noah rambling… probably

A little well-kept secret for companies crushing it with AI is that only a handful of AI tools are actually worth using.

That’s why we—Noah & Grant—use ChatGPT and Claude, along with a killer product called Attention.

Here’s how it works:

We tell Attention exactly what to pay attention to in our meeting (goals, budget, etc).
Attention listens in to our call.
After, Attention outputs important insights, action items, and follow-up emails.

There are way more features, and it’s a must-have for all sales teams. (If you sell anything, literally drop what you’re doing and go book a demo with Attention ASAP).

Prompt Tip of the Day

Greg Brockman shared the following prompt format for o1 (OpenAI’s reasoning models): Goal → Return Format → Warnings → Context Dump.

The game-changer? That context dump. “We won't see each other for weeks, so uniqueness really counts” hits different than just "unique places please."

Try it: Next time you're stuck with generic AI responses, dump in the context you'd share with a friend. You can even upload this screenshot and say “help me write a prompt with this structure for the below context.”

Treats To Try.

Career Dreamer from Google helps you discover new career paths by analyzing your life experiences and showing how your existing skills match different jobs (read more).
Proxy takes over repetitive web tasks for you, from checking your morning news to sending LinkedIn messages—clicking, scrolling and navigating websites without APIs (free to start, $20 a month for more scheduled automations).
Perplexity released its own custom version of DeepSeek R1 (a “finetune” called R1 1776 lol) that removes the original’s censorship (read more).
Photoroom removes backgrounds and edits product photos in seconds (free and paid versions).
Brainfish watches how you use your product, automatically creates and updates help docs, and provides real-time assistance to customers when they get stuck—free to try.
Octopus can create visual sitemaps instantly by crawling your website’s pages and links.
Midlife Engineering lets you craft calming sound environments to destress and focus, using a digital music mixer (just click around on the keyboard thing).

See our top 51 AI Tools for Business here!

Around the Horn.

Claude beat both OpenAI’s 4o and o1 models at coding on OpenAI’s own SWE-Lancer benchmark that measures how good AI is at replacing software devs.
Apple’s new budget iPhone, the iPhone 16E, comes with Apple Intelligence for only $599—it can do that because it uses the same chip as the iPhone 16.
Google Lens now analyzes and explains anything you see on your iPhone screen, instead of just finding matching images online.
Microsoft created a quantum chip called Majorana 1 using a new type of material that makes quantum computing more stable and scalable.
Rabbit previewed its new Android agent that uses your phone apps to send messages, create documents, and manage tasks for you (video).

FROM OUR PARTNERS

🥋Master your unstructured GenAI data

Join Unstructured & Elastic for a live webinar on Feb 25 to accelerate your learning curve on the latest techniques for fueling your GenAI apps and pipelines with all your document data.

Click here to register and receive the recording, even if you can’t attend live

Thursday Trivia

One is real, and one is AI. Which is which? (vote below!)

Which is AI?

The answer is below, but place your vote to see how your guess compares to everyone else (no cheating now!)

Here are the results from last week’s trivia (A was AI):

Y’all smelled this one a mile away—9 humans, 4 robots.

Here’s what you said:

L.B. chose A: “The lack of letters and the weird hand are a dead giveaway. It's easy when you know what to look for! :D”
C.N. chose B: “I think B is AI because the figure doesn't have any slouch in his posture!”
A.S. chose A: “Because, you know, two computers back-to-back at a convenience store is totally normal.”