Welcome back! A Tom's Guide writer challenged ChatGPT to make everyday chores more fun, and apparently the secret is turning adulthood into a video game. With a little help from plugins like Todoist and Spotify, cleaning became a timed challenge with experience points, laundry got its own soundtrack, and decluttering turned into a series of mini-missions. Finally, a way to convince ourselves that doing the dishes is character development.
World Models Just Went Open-Source
Last week, Google's Genie 4 showed how far AI-generated worlds have come. This week, I found something that might matter even more.
I shared an open-source world model that does a lot of what Genie 4 does, and the big difference is you can actually try it right now:
Interactive environments generated on the fly, not just video clips you watch.
Open-source so developers can run it, study it, and build on top of it.
Available to test today, instead of sitting behind a research preview.
My POV: The closed labs still set the pace, but open-source keeps closing the gap faster than anyone expects. We saw it with language models, then image and video models, and now it's happening with world models. Once anyone can tinker with this stuff, the most interesting uses usually come from people nobody was paying attention to.
Hit reply and let me know: would you rather have the most polished model behind a waitlist, or a slightly rougher one you can use today?
— Matt
Anthropic Reveals Eye-Watering Numbers Ahead of IPO
Via CNBC
Anthropic disclosed new financial details in an updated S-1 filing ahead of its highly anticipated public debut, revealing enormous revenue growth alongside nearly $900 billion in infrastructure commitments.
By the numbers:
$17 billion in Q3 revenue, up from $1.4 billion a year earlier.
47% quarterly revenue growth, down from 144% in the previous quarter.
$23.3 billion net loss, narrowed from $66.3 billion in Q2.
$895.4 billion in purchase and lease commitments, almost entirely related to computing infrastructure, with approximately $69.8 billion due over the next 12 months.
The infrastructure gamble: Anthropic is committing extraordinary sums to the computing power required to train and run Claude. Its filing warns that some infrastructure partners rely on outside financing, meaning delays or higher costs could affect the capacity Anthropic has contracted for.
The bigger picture: These numbers are enormous, even by AI industry standards. Anthropic is generating revenue at an astonishing pace, but its spending commitments show just how expensive staying at the frontier has become. As the company approaches its stock market debut, investors will have to decide whether that growth can continue long enough to justify the valuation, or whether slowing momentum and mounting infrastructure costs will eventually catch up.
DeepSeek Takes Aim at Claude Code
Via Financial Post
Chinese AI startup DeepSeek released a prototype of its new open-source coding assistant, Super-Agent, designed to challenge Anthropic's Claude Code.
What it does: Unlike traditional coding assistants that wait for a prompt, Super-Agent can proactively monitor developers' group chats, identify coding problems, and suggest solutions. The idea is to make AI less of a tool developers constantly supervise and more of a teammate that notices when something needs attention.
A crowded field: Claude Code, OpenAI Codex, Cursor, and a growing list of coding agents are already competing for developers' attention. DeepSeek's open-source approach could appeal to teams seeking more control over their tools, although this is still a prototype, so its performance against established competitors remains to be seen.
The bigger picture: Coding has become one of the most competitive and commercially promising areas of AI. Nearly every major lab is racing to build assistants that don't just suggest code but actually complete development tasks. DeepSeek's entry shows how quickly the market is evolving, with the next battleground increasingly focused on autonomy, workflow integration, and how little developers need to intervene.
Granola just got a lot crunchier
Granola, the AI notepad for people with back-to-back meetings, just announced a voice-to-text feature.
You can now dictate words in any app. Because Granola has been in your meetings, it already knows your colleagues' names and your jargon, so the spelling is right the first time around. Unlimited words for all Granola users.
Try it here. 1 month off any paid plan with code MATTWOLFE.
Turn Company Documents Into a Searchable Wiki
Via Ophiry
Ophiry transforms company documents, manuals, policies, and handbooks into structured, sourced wiki articles. Its AI Editor-in-Chief organizes the information, while a separate assistant answers employee or client questions with citations to the original materials.
How you can use it
Convert internal documents into organized wiki articles
Answer questions with company knowledge
Find information and trace citations to original sources
Control access by user, team, or client
Keep company information on EU-hosted servers
Pricing: Paid, free trial available
Animate Characters From a Single Image
Via Motion Control AI
Motion Control AI is a browser-based video generation tool that transfers movement from a reference clip onto an uploaded character image. Powered by Kling 3.0 Motion Control, it recreates gestures, facial expressions, body movements, and camera motion without requiring a traditional studio or motion-capture setup.
How you can use it
Animate characters using a single image
Transfer movements from existing videos
Choose from 36 built-in reference motions
Generate character videos in vertical, widescreen, or square formats
Export videos in 720p or 1080p
Pricing: Paid
Jobs, announcements, and big ideas
OpenAI shows how Codex and Astra are speeding up hardware and CAD engineering.
Apple Watch gains Siri recaps and a "Live Rewind" for audio in the latest watchOS beta.
Google brings Guided Learning to Classroom and tightens Gemini controls for K-12.
Netflix adds an "AI Discover" tab that serves up themed picks on TV.
OpenAI renames its data centers "compute campuses" as local pushback grows.
xAI launches Data Bot to give any team an on-call data scientist.
Claude's API cuts agent latency by 35% with new message threads.
Generate worlds at your fingertips: This might be the coolest thing you see this week. Follow along below to try!
That’s a wrap! See you this Friday for more.
