AI grows a brain

Is Meta’s new model Isaac Newton?

June 13, 2025

Welcome back! At its WWDC extravaganza earlier this week, Apple forgot to invite one guest to the party: Siri.

While the company announced Apple Intelligence and unveiled new features like live translation and ChatGPT-powered image edits, its own AI assistant barely made an appearance. The most we got? A promise that Siri updates were still “in progress.”

That silence is loud. Google and Microsoft are already shipping assistants that see your screen, summarize your files, and act on your behalf. But Apple is still explaining why last year’s Siri upgrades are missing this year’s keynote. What do you think—what’s Apple’s play here?

OpenAI’s o3-pro thinks better than you do

TechCrunch

This week, OpenAI released o3-pro, the latest version of its “reasoning-first” model architecture—and it might be the best glimpse yet at what comes after GPT-4.

While GPT-4o focused on multimodal flair, o3-pro doubles down on intelligence, logic, and nuance. Sam Altman even teased that AI will start producing novel insights next year. With models like this? It’s believable.

What makes o3-pro different:

  • This model is designed for performance under pressure. It improves significantly on tricky reasoning tasks, especially those with ambiguous or multi-step logic.

  • In tests across math, science, and coding problems, o3-pro showed stronger consistency and planning than its predecessors—even when hallucinations and flukes were stripped away.

Why it matters: Where other models might succeed by accident, o3-pro is built to succeed reliably and with stability. OpenAI’s new evals focus on whether models can deliver “guaranteed correct” answers, not just cherry-picked examples. And o3-pro passes that bar often.

Meta thinks your next AI assistant should understand gravity

Meta dropped a new model designed to help AI agents make sense of the 3D world around them. Built on 100,000 hours of simulated and real-world egocentric video, it’s trained to perceive depth, physics, object permanence, and even what might be just out of view. 

A model that can see, plan, and act: The model, dubbed “World Model,” ingests raw video and turns it into a structured representation of physical environments complete with 3D meshes, segmentation, and predicted movements.

  • Meta says it enables agents to perform tasks like grabbing objects from under a table, choosing routes around obstacles, or predicting what happens if something falls.

  • These are basic to humans, but still a leap for AI.

The bigger trend: AI’s next frontier is physical. This is Meta's clearest move yet into robotics and embodied AI, and it’s betting not just on what AI can say, but what it can do in the real world. Would you trust a robot that can anticipate your next move?

Hollywood strikes back

Major studios are taking their first legal swing at generative AI.

Disney and NBCUniversal filed a joint lawsuit against Midjourney, accusing the AI image generator of copyright infringement for allegedly using their characters (like Simba and the Minions) to train its model and generate lookalikes.

Studios aren’t playing around: The complaint claims the studios attempted to resolve the issue privately, but Midjourney “ignored” their demands. Instead, it allegedly kept releasing new versions of its model that generated even more “higher quality infringing images.”

This isn’t isolated: In February, over a dozen news companies filed suit against Cohere for similar IP violations. The lawsuits are increasingly targeting the platforms themselves, not just users.

The bigger trend: Hollywood is going after AI tools that recreate copyrighted characters pixel-for-pixel. If successful, this case might set the precedent for how generative AI can legally train and create with copyrighted content.

I'm curious: Do you think Hollywood and AI can have a positive relationship?

Click one and tell me more of your POV

Login or Subscribe to participate in polls.

One workspace, infinite research agents

Skywork

Skywork AI tackles deep research and content creation across documents, slides, spreadsheets, webpages, and even podcasts. The product is powered by its proprietary DeepResearch engine (ranked #1 on the GAIA benchmark).

How you can use it:

  • Turn dense topics into fully cited, well-formatted reports

  • Instantly convert docs into podcasts or presentations

  • Benchmark competitive landscapes and market data across industries

Pricing: Free and paid plans available

Your GitHub issues, solved in 3 minutes

GitAuto

GitAuto is an AI-powered agent that connects directly to your GitHub or Jira backlog and autonomously generates, tests, and submits code for your feature requests or bug fixes.

How you can use it:

  • Autogenerate fixes for issues or Jira tickets

  • Review and merge production-ready PRs

  • Reduce engineering costs and backlogs

Pricing: Free plan available with team plan starting at $100/month

Jobs, announcements, and big ideas

  • Dia is making a bold bet on the web and placing an even bigger one on AI.

  • Scale AI and Alexandr Wang are trying to reignite Meta’s push into AI.

  • The state of New York is tracking whether AI causes layoffs.

  • Join AWS as an Applied Scientist to develop core natural language processing, agents, generative AI, deep learning, and machine learning algorithms.

  • Meta launched generative AI video editing tools across all of its apps.

Can AI unlock the fountain of youth? I sat down with Superpower’s founder to learn how AI could monitor your body 24/7 and help you live to 120.

That’s a wrap! See you next week.

—Matt (FutureTools.io)

P.S. This newsletter is 100% written by a human. Okay, maybe 96%.