AI Action Plan
The Trump Administration unveiled its much-awaited AI Action Plan, a set of priorities, directives, and policies geared towards helping America "achieve global dominance in artificial intelligence."
Between the lines:
The Office of Science and Technology Policy drafted the plan focusing on three key “pillars”: accelerating AI innovation, building infrastructure, and leading international diplomacy and security.
With over 90 policies recommended, it’s a wide-ranging document that, in practice, aims to roll back regulatory hurdles in developing new AI models - in short, much of what Silicon Valley has been asking for.
The strategy represents a more aggressive approach to AI development, emphasizing expanding energy supplies for data centers and other AI infrastructure.
And many headlines are already being written about the Action Plan's commitment to fighting AI bias, not from a safety perspective, but from an ideological one. See: the White House’s executive order fighting "woke AI" in the federal government.
Elsewhere in AI regulation:
Over $1 billion worth of Nvidia's AI chips were shipped to China three months after President Trump tightened chip export controls in May.
After companies criticized the lack of clarity, the EU issued guidelines for AI compliance to help AI models with systemic risks meet AI Act requirements.
Over 500 organizations lobbied on AI in the first half of 2025, with OpenAI spending $1.8 million, up from $380,000 in 2023.
And Meta announced it won't sign the EU's AI code of practice, calling the guidelines overreach that will throttle AI development and deployment in Europe.
Gold Standard
Both DeepMind and OpenAI achieved a gold medal performance on the International Mathematical Olympiad (IMO). While both AI labs achieved the same score (35 out of 42 points), DeepMind underwent a more formal, rigorous process.
Why it matters:
While it may seem like a somewhat niche achievement, the IMO is perhaps the most prestigious competition for young mathematics students worldwide - and another domain where AI has achieved elite human performance.
What’s perhaps more intimidating is how fast it got there. DeepMind scored a silver medal last year, but did so using a “formal” system - meaning the problems had to be translated into a machine-readable syntax. Now, both labs used "informal" systems with natural language inputs.
Aside from being “experimental reasoning models,” details on the architectures are scant - but the hope is that the models behind this result will be far more capable of solving large-scale, complex problems than the ones we have today.
Elsewhere in frontier models:
Alibaba released its updated Qwen3 model, Qwen3 Coder for agentic coding, and open-sourced Qwen Code, a CLI tool adapted from Gemini CLI.
Google DeepMind unveiled Aeneas, an AI model for contextualizing ancient Latin inscriptions. The model will help historians interpret, attribute, and restore text fragments.
And Anthropic and other researchers detail "subliminal learning", where LLMs learn traits from model-generated data that is semantically unrelated to those traits.
Elsewhere in Google's launch calendar:
Google launched Web Guide, an AI-powered Search Labs experiment that organizes Google Search results by grouping pages related to specific aspects of the query.
The company rolled out its virtual clothes try-on feature, which lets users upload photos of themselves, across the US.
YouTube gives Shorts creators access to new generative AI features powered by Veo 2, such as new AI effects and a tool that turns photos into six-second videos.
And Google is seeking to recruit news organizations for a new licensing project related to AI, and plans to launch a pilot initially with ~20 US outlets.
The Eternaut
For the first time, Netflix has used generative AI visual effects in an original production, creating a building collapse sequence in "The Eternaut" at 10x the speed and a fraction of the cost of traditional methods.
The big picture:
While it's unclear what technology Netflix used for this film, other reports suggest that the company uses Runway's AI video generation tools elsewhere.
It's the first major streaming platform to deploy generative AI in final footage publicly, signaling broader industry acceptance despite ongoing controversy over job displacement and content ownership (i.e., the writer's strike two years ago).
The technology is also reshaping the world of short-form social video, as creators like the Dor Brothers become content creation powerhouses with hundreds of millions of views and millions in revenue - all through fully AI-generated video content.
And we've barely scratched the surface: Veo 3 can generate videos with realistic sound, while Decart's Mirage can transform live video in real time, a game changer for live streaming and content creation.
Elsewhere in AI anxiety:
The US FDA's AI tool Elsa has fabricated nonexistent studies, misrepresented research, and cannot access relevant documents to assist with review work.
Dozens of xAI staff voiced concerns over a program granting xAI "perpetual" access to data like their "likeness" for training, with several not consenting.
LeBron James' lawyers sent a cease-and-desist to an AI company making pregnant videos of him.
Spotify published AI-generated songs from dead artists without permission.
The NIH is capping research proposals because it's overwhelmed by AI submissions.
And Pinterest is increasingly filled with AI interior design images, often featuring unrealistic or impossible details that complicate users' search for inspiration.
Things happen
OpenAI plans to launch GPT-5 in early August Q&A with Stability AI's new CEO Prem Akkaraju Death by AI AI comes up with bizarre physics experiments, but they work.Proton launches Lumo, an AI chatbot with "zero-access" encyption. Nobody knows how to build with AI yet. How AI is reshaping higher education. Scale AI is replacing low-cost data labelers with highly paid experts. AI capex is so big it's affecting economic statistics. 75% of US teens have used AI as a companion. Mistral releases the first environmental lifecycle analysis of an AI model. Stop building AI tools backwards. Google users are less likely to click links when AI summaries appear. Nvidia rival Hailo launches its second-generation AI chip. It's rude to show AI output to people. Companies like McDonald's are turning to India for AI talent. Swedish AI startup Lovable became the fastest-growing software startup ever. How I keep up with AI progress. ChatGPT hallucinated a feature, forcing developers to add it. Walmart consolidates its AI into four super agents. Two narratives about AI. Big tech companies intensify efforts to reduce AI hallucinations. AMD unveils the first Stable Diffusion 3.0 model for Ryzen laptops. AI market clarity. Google's AI is destroying search, the internet, and your brain. Satya Nadella addresses the "enigma" of layoffs while Microsoft thrives. Exhausted man defeats AI in coding championship. OpenAI and UK sign strategic partnership. Nvidia's CUDA now supports RISC-V. Why are we pretending AI is going to take all the jobs? Hackers plant computer wiping commands in Amazon's AI coding agent. Walmart hires Instacart's CPO to lead AI efforts. The big LLM architecture comparison. How open-source helped Chinese AI companies take on Silicon Valley. OpenAI's $40B funding round is reopening. Grok's revenue jumped 325% after launching Grok 4. Oracle to supply OpenAI with 2M AI chips. Google's Windsurf deal is a wake-up call for AI startup employees. Mark Zuckerberg offers OpenAI researchers $300M packages. Figma makes its prompt-to-app tool available to all users. UMG partners with Liquidax Capital to develop music AI patents. Microsoft's AI chief focuses on "humanist superintelligence". ChatGPT users send 2.5B prompts daily. SoftBank and OpenAI disagree on Stargate terms. Apple losing ground in AI talent war. Amazon shuts down its Shanghai AI lab. OpenAI's Fidji Simo lays out vision for AI. Amazon acquires Bee, a $50 always-listening AI wearable. Google's AI Overviews have 2B monthly users. OpenAI is building an office productivity suite. Microsoft has poached 24+ employees from Google DeepMind.
