AI develops human-like object understanding
PLUS: OpenAI and Microsoft tensions are reaching a boiling point
Howdy. This is the “AI Daily” newsletter, your daily dose of AI news and resources.
In today’s edition:
$2,000 fully AI-generated ad that aired during the NBA finals
Alexandr Wang leaves Scale AI to lead Meta’s Superintelligence
Sam Altman thinks AI will have ‘novel insights’ next year
Trump administration's whole-government AI plans leaked on GitHub
Plus trending AI tools, posts, and resources
Ready, set, go…
$2,000 fully AI-generated ad that aired during the NBA finals
In a groundbreaking moment for advertising and artificial intelligence, the 2025 NBA Finals featured the first-ever fully AI-generated commercial aired during a major primetime sports event.
Here's what you need to know:
Directed by self-described “AI filmmaker” PJ Accetturo, the ad was produced entirely using Google’s Veo 3, a text-to-video AI generator.
He utilized AI tools like Gemini and ChatGPT to develop the script, shot list, and prompts, resulting in 15 usable clips from 300–400 generations.
The entire production was completed in just two days at a cost of approximately $2,000, a stark contrast to traditional commercial production budgets that often reach six to seven figures.
Created for the prediction market platform Kalshi, used surreal and unconventional imagery, including a farmer floating in a pool of eggs, an alien chugging beer, and an elderly man draped in an American flag exclaiming, “Indiana gonna win baby.”
Why it matters:
As AI technology continues to evolve, the Kalshi ad serves as a compelling example of how artificial intelligence is going to revolutionize content creation, offering new possibilities for storytelling and audience engagement in the advertising industry.
AI video is now good enough for primetime ads, offering brands a way to create eye-catching (if surreal) content at a fraction of traditional costs. While some will still pay for "premium" human-crafted spots, the rise of tools like Veo 3 means weird, hyper-engaging AI ads will dominate 2025.
AI develops human-like object understanding
New research revealed that AI models are spontaneously developing internal 'maps' of the world that closely mirror human conceptual understanding, making a leap from simple recognition to machine cognition.
Here's what you need to know:
AI models were tested on 4.7M "odd-one-out" decisions across nearly 2,000 common objects, studying how they organize and understand the world.
The AI naturally developed 66 core ways of thinking about objects, closely matching how humans mentally categorize things like animals, tools, and food.
The AI’s conceptual map showed a strong alignment with human brain activity patterns, particularly in regions responsible for processing object categories.
Rather than just memorizing patterns, the research showed that AI models build genuine internal concepts and meanings for objects.
Why it matters:
While there is still a large subset of the AI world that believes that models are just “stochastic parrots”, we’re continuing to see more and more evidence of the tech showing hints of genuine reasoning and conceptual understanding—and that machine intelligence may work in similar ways to humans after all.
Side Updates
Disney and Universal sue Midjourney
Disney and Universal have filed a major lawsuit against Midjourney, accusing the AI company of mass copyright infringement by generating images of iconic characters like Darth Vader, Elsa, Iron Man, and Shrek. The suit, filed in California, slams Midjourney’s tool as a “virtual vending machine” that illegally churns out and markets Disney and Universal IP without permission or payment. With Midjourney reportedly training a video generator, the studios fear future violations are imminent. They're pushing for a jury trial, calling it a clear-cut case of infringement, marking the first major Hollywood legal battle against generative AI.
Alexandr Wang leaves Scale AI to lead Meta’s Superintelligence
Alexandr Wang is stepping down as CEO of Scale AI to join Meta and lead its superintelligence efforts, following Meta’s massive $14.3 billion investment in the startup. Although Meta now holds a 49% non-voting stake in Scale AI, the company says this won’t affect existing clients like Google, Microsoft, or OpenAI. Jason Droege, former Uber VP and current chief strategy officer, will take over as CEO. A few Scale employees will also move to Meta. This move reflects Mark Zuckerberg’s focus on advancing Meta’s AI capabilities after the lukewarm reception of its latest Llama model.
Sam Altman thinks AI will have ‘novel insights’ next year
OpenAI CEO Sam Altman’s new blog post, “The Gentle Singularity,” claims we’ve already entered the AI singularity, but instead of a chaotic takeover, it feels more like upgrading internet speed. He says the hardest scientific work is done, and what's left is scaling. Altman envisions ultra-personalized AI, where “idea people” thrive and intelligence becomes nearly free. His subtle roadmap predicts agentic AI by 2025, AI scientists by 2026, and humanoid robotics by 2027. But he acknowledges big risks remain, including alignment, regulation, and infrastructure limits. If those fail, his gentle vision could still turn turbulent.
OpenAI and Microsoft tensions are reaching a boiling point
OpenAI wants to loosen Microsoft's grip on its products and computing resources and secure approval for its conversion into a for-profit company, but negotiations have been difficult. The company's executives have discussed the option of accusing Microsoft of anticompetitive behavior during their partnership - this could result in a regulatory review of the terms of the contract for potential violations of antitrust law. OpenAI must complete its conversion to a public-benefit corporation by the end of the year or it risks losing $20 billion in funding.
The US government is considering not regulating AI for 10 years
The U.S. government is considering a new law that would block individual states from making their own rules about AI for the next 10 years. This rule is hidden inside a big bill recently passed by the House of Representatives, and now it's going to the Senate. If approved, it would cancel more than 60 state AI laws and stop others from being made. Supporters like Meta and OpenAI say this will help companies grow faster and avoid confusing laws. But critics say it gives too much power to tech companies and takes away important protections for people.
Trump administration's whole-government AI plans leaked on GitHub
The Trump administration is quietly preparing to launch a new government-wide AI initiative on July 4, based on a now-hidden GitHub repository tied to the “ai.gov” website. Spearheaded by former Tesla engineer Thomas Shedd, the project aims to make the U.S. government operate like a tech startup, using AI to automate federal work. The site will reportedly include a chatbot, an “all-in-one API” for connecting to OpenAI, Google, and Anthropic models, and a dashboard called CONSOLE to monitor usage. While vendors are mostly FedRAMP certified, concerns remain over data security and rapid AI integration without clear oversight.
Meta launches AI world model to advance robotics and self-driving cars
Meta has unveiled V-JEPA 2, a new open-source AI “world model” designed to understand the 3D physical world, including how objects move and interact. Unlike traditional AI that relies on labeled data, V-JEPA 2 uses simplified internal simulations to reason and predict outcomes, like knowing a ball will fall if it rolls off a table. Meta says this could power smarter delivery robots and self-driving cars. The announcement comes as Meta doubles down on AI, planning a $14 billion investment in Scale AI and reportedly bringing on its CEO, Alexandr Wang, to accelerate its broader AI strategy.
Useful AI Links
Trending Tools
Dream Recorder: The magical bedside open-source device that plays your dreams back as cinematic reels.
Underlord by Descript: An AI video editor for vibe editing.
Factory AI: Delegate software development tasks to agents called Droids. Droids take commands and deliver: pull requests, tickets, docs, and more.
Perplexity Finance: Download Excel models directly from finance pages for seamless financial modeling.
Dia: The AI-powered browser assistant that enhances your browsing experience by helping you write, learn, plan, and shop by using context from the pages you're visiting.
Nourri AI: AI that takes the guesswork out of nutrition, understanding your plate to offer personalized insights.
Chat4Data: Think of it as ChatGPT for web scraping, effortlessly extracting and processing data from any website.
Resources / Guides
How I use LLMs.
AI has fundamentally made me a different person.
Don't sell shovels, sell treasure maps.
Sam Altman explains what’s coming after GPT-4.
What is ChatGPT doing, and why does it work?
A prompt that writes your Entire Business Plan in minutes.
A 29-videos playlist on how to build DeepSeek from scratch. It covers theory and code, from the very foundations to advanced.
After 6 months of daily AI pair programming, here's what actually works.
Why DeepSeek models are good at reasoning.
Reddit was worthless until ChatGPT.
AI can now stalk you with just a single vacation photo.
News sites are getting crushed by Google’s new AI tools.
Build AI collaboration skills with Anthropic’s hands-on fluency course.
Model Context Protocol (MCP) course by Huggingface.
He automated 73% of his remote job using basic automation tools, told his manager everything, and got a promotion.
Anthropic released a blog sharing the practical lessons they learned while building a multi-agent research system.
Run autonomous browser tasks with Gemini 2.5.
Jobs Board
Daily Dose of Contents
1/
The state of vibe coding in June 2025.
2/
I love these kind of smart takes with perspective. I feel like most people’s AI’s takes have no understanding of how the world or people actually work and this right here is the context I’ve been grasping at.
3/
A humanoid robot, Nadia, is remotely controlled for boxing training using a simple VR motion capture setup.
That’s All For Today, Folks
Thanks again for being here.