The Rundown AI / Articles / AI / Thinking Machines co-founder heads to Meta

Thinking Machines co-founder heads to Meta

PLUS: Patients control AI and robotics with thought

Zach Mink

October 13, 2025

Good morning, AI enthusiasts. Leaving a $12B startup you founded right after shipping its first product (and after already rejecting a rumored $1.5B offer) means the counter must be absolutely insane.

Mark Zuckerberg’s successful poaching of Mira Murati's Thinking Machines Lab co-founder just showed Meta's 'hiring freeze' thaws very quickly when it comes to the top AI talent.

Want to join The Rundown? We’re looking to fill seven new positions, and are offering a $2000 referral bonus for full-time hires. View the openings here.

In today’s AI rundown:

Thinking Machines Lab co-founder joins Meta
xAI’s world models for video game generation
Connect Agent Builder to 8,000+ tools
OpenAI’s GPT-5 reduces political bias by 30%
4 new AI tools, community workflows, and more

LATEST DEVELOPMENTS

TALENT WARS

👀 Thinking Machines Lab co-founder joins Meta

Image source: Reve / The Rundown

The Rundown: Andrew Tulloch, the co-founder of Mira Murati’s Thinking Machine Lab, just departed the AI startup to rejoin Meta, according to the Wall Street Journal, marking another major talent acquisition for Mark Zuckerberg’s Superintelligence Lab.

The details:

Tulloch spent 11 years at Meta before joining OpenAI, and reportedly confirmed his exit in an internal message citing personal reasons for the move.
The researcher helped launch Thinking Machines alongside former OpenAI CTO Mira Murati in February, raising $2B and building a 30-person team.
Meta reportedly pursued Tulloch this summer with a compensation package as high as $1.5B over 6 years, though the tech giant disputed the numbers.
The hiring comes as Meta continues to reorganize AI teams under its MSL division, while planning up to $72B in infrastructure spending this year.

Why it matters: TML recently released its first product, and given that Tulloch had already reportedly turned down a massive offer, the timing of this move is interesting. Meta’s internal shakeup hasn’t been without growing pains, but a huge infusion of talent, coupled with its compute, makes its next model a hotly anticipated release.

TOGETHER WITH NEBIUS

📶 Why AI teams are switching from hyperscalers to Nebius

The Rundown: Hyperscalers promise AI-ready infrastructure, but deliver generic cloud with GPU bolt-ons. Nebius designed their platform from the ground up for machine learning workloads, with the performance guarantees and developer experience that AI engineers actually need.

By switching to Nebius, you receive:

3 months free service plus zero-cost data migration from your current provider
Supercomputer performance with cloud-like flexibility and familiar tooling
Solution architects who guide you through every step of the transition

Learn more & switch today.

XAI

🎮 xAI’s world models for video game generation

Image source: Reve / The Rundown

The Rundown: Elon Musk's xAI reportedly recruited Nvidia specialists to develop world models that can generate interactive 3D gaming environments, targeting a playable AI-created game release before 2026.

The details:

xAI hired Nvidia researchers Zeeshan Patel and Ethan He this summer to lead the development of AI that understands physics and object interactions.
The company is recruiting for positions to join its “omni team”, and also recently posted a ‘video games tutor’ opening to train Grok on game design.
Musk posted that xAI will release a “great AI-generated game before the end of next year,” also previously indicating the goal would be a AAA quality title.

Why it matters: World models have been all the rage this year, and it’s no surprise to see xAI taking that route, given Musk’s affinity for gaming and desire for an AI studio. We’ve seen models like Genie 3 break new ground in playable environments — but intuitive game logic and control are still needed for a zero-to-one gaming moment.

AI TRAINING

🔌 Connect Agent Builder to 8,000+ tools

The Rundown: In this tutorial, you will learn how to connect OpenAI's Agent Builder to over 8,000 apps using Zapier MCP, enabling you to build powerful automations like creating Google Forms directly through AI agents.

Step-by-step:

Go to platform.openai.com/agent-builder, click Create, and configure your agent with instructions like: "You are a helpful assistant that helps me create a Google Form to gather feedback on our weekly workshops." Then select MCP Server → Third-Party Servers → Zapier
Visit mcp.zapier.com/mcpservers, click "New MCP Server," choose OpenAI as the client, name your server, and add apps needed (like Google Forms)
Copy your OpenAI Secret API Key from Zapier MCP's Connect section and paste it into Agent Builder's connection field, then click Connect and select "No Approval Required"
Verify your OpenAI organization, then click Preview and test with: "Create a Google Form with three questions to gather feedback on our weekly university workshops." Once confirmed working, click Publish and name your automation

Pro tip: Experiment with different Zapier tools to expand your automation capabilities. Each new integration adds potential for custom workflows and more advanced tasks.

PRESENTED BY ATLASSIAN

🎯 What the 4% of AI-successful companies do differently

The Rundown: AI usage has doubled in the past year, and workers report being 33% more productive. So why haven't isolated productivity gains translated to company-wide transformation?

Atlassian's AI Collaboration Index reveals:

Most teams remain buried in tasks and disconnected across applications
Personal efficiency is up, but AI hasn't solved the team-wide collaboration problem
Winners are focused on company-wide systems for AI coordination, not individual tools

Download the AI Collaboration Index to learn what sets top performers apart.

AI RESEARCH

📊 OpenAI’s GPT-5 reduces political bias by 30%

Image source: OpenAI

The Rundown: OpenAI just released new research showing that its GPT-5 models exhibit 30% lower political bias than previous models, based on tests using 500 prompts across politically charged topics and conversations.

The details:

Researchers tested models with prompts ranging from "liberal charged" to "conservative charged" across 100 topics, grading responses on 5 bias metrics.
GPT-5 performed best with emotionally loaded questions, though strongly liberal prompts triggered more bias than conservative ones across all models.
OpenAI estimated that fewer than 0.01% of actual ChatGPT conversations display political bias, based on applying the evaluation to real user traffic.
OAI found three primary bias patterns: models stating political views as their own, emphasizing single perspectives, or amplifying users’ emotional framing.

Why it matters: With millions consulting ChatGPT and other models, even subtle biases can compound into a major influence over world views. OAI’s evaluation shows progress, but bias in response to strong political prompts feels like the exact moment when someone is vulnerable to having their perspectives shaped or reinforced.

QUICK HITS

🛠️ Trending AI Tools

🎥 Sora 2 - OpenAI’s SOTA video model, now available via API
⚙️ Claude Code - Anthropic’s AI coding assistant
💼 Gemini Enterprise - Discover, create, share, and run AI agents
📱 Apps SDK - Chat with and build apps directly in ChatGPT

📰 Everything else in AI today

Atlassian announced the GA of Rovo Dev. The context-aware AI agent supports professional devs across the SDLC, from code gen and review to docs and maintenance. Explore now.*

OpenAI served subpoenas to Encode and The Midas Project, demanding communications about California's AI law SB 53, with recipients calling it intimidation.

Apple is reportedly nearing an acquisition of computer vision startup Prompt AI, with the 11-person team and tech set to be incorporated into its smart home division.

Several models achieved gold medal performance at the International Olympiad on Astronomy & Astrophysics, with GPT-5 and Gemini 2.5 receiving top marks.

Mark Cuban opened up his Cameo to public use on Sora, using the platform as a tool to promote his Cost Plus Drugs company by requiring each output to feature the brand.

Former UK Prime Minister Rishi Sunak joined Microsoft and Anthropic as a part-time advisor, where he will provide “strategic perspectives on geopolitical trends”.

*Sponsored Listing

COMMUNITY

🤝 Community AI workflows

Every newsletter, we showcase how a reader is using AI to work smarter, save time, or make life easier.

Today’s workflow comes from reader John C. in Seattle, WA:

"I’m a first-time founder building a product for restaurants. I…wanted to feel like I was working with a business partner. I built a custom GPT with the persona of Steve Jobs (my idol), who is a veteran food industry exec instructed to share advice. The responses I get are informed, no-BS, and sprinkle in delightful quotes from Jobs and fictional but relevant anecdotes from previous food industry experience.”

How do you use AI? Tell us here.