👋 Hello hello,
OpenAI is out here solving math problems that have stumped actual mathematicians for over a decade, while the music industry is trying to keep AI songs off the charts entirely.
Also, the EU just made it a lot harder to sneak a deepfake past anyone without a label on it.
Let’s get into it.
🔥🔥🔥 Three Curated AI Updates
An internal version of OpenAI's next major model produced new results on ten long-standing open problems in mathematics and theoretical computer science. We already told you about this yesterday and OpenAI just published a blog on this as well.
So if you’re asking yourself why this matters, here’s why: math like this runs underneath things we all touch, GPS, weather forecasting, medical imaging, how diseases spread. Progress at the foundational level tends to ripple outward eventually. Solving decades old problems could mean breakthroughs across technologies. OpenAI is releasing the manuscripts, formal Lean certificates, and even the model's own reasoning walkthroughs, so mathematicians can dig in and build on the ideas themselves.
Illustration by Neil Jamieson
A coalition of major and independent record labels, including Universal Music Group, Sony, and Warner alongside BMG and Concord, proposed new rules for when an AI-involved track can qualify for official charts. Under the proposal, a song only counts if it's "substantially human-made," made using AI tools that are properly licensed, and free of streaming manipulation.
It looks contradictory on paper, since some of these same labels are actively partnering with AI companies like Suno, Udio, and Nvidia while pushing this proposal. The real goal here though is control: labels want to set the terms for how AI shows up in music, instead of waiting for regulation to catch up on its own.
Under Article 50 of the EU AI Act, companies now have to make AI-generated and AI-manipulated content detectable. Providers like OpenAI and Anthropic need to mark their outputs in a machine-readable way, and anyone deploying a realistic deepfake has to disclose it clearly.
The rules zero in on realistic content specifically. A photorealistic deepfake of a public figure needs a label, an obviously fantastical AI image of a dragon does not. AI-generated text on matters of public interest needs disclosure too, unless a human has actually reviewed and edited it.
It's the EU's biggest swing yet at making synthetic content identifiable at scale, and it puts real pressure on platforms to build detection into their pipelines, not just their policies. Whether enforcement can keep pace with how fast this content spreads is still an open question.
🔥🔥 Two Pro AI Tools
1.🛡️ Balance Theory
Balance Theory is an AI-native platform built for CISOs to manage cybersecurity investment decisions instead of just buying more tools. It currently oversees more than $1 billion in security spend and just closed a $19M round to build out its AI layer further. Best for security leaders who need to justify budget and actually see what their stack is doing.
2. 🎙️ Smallest.ai
Smallest.ai builds real-time voice AI, ultra-low latency text-to-speech and conversational voice agents that respond in well under a second, support 30+ languages, and can clone a voice on demand. Best for teams building customer support bots or voice products who are tired of AI voices that sound like AI voices.
🔥 Things You Didn’t Know You Could Do With AI
A developer wanted to send a file between two phones that weren't on the same network, so they built a way to do it using nothing but a screen and a camera, with Claude Code putting the prototype together overnight.
Split the file into small blocks and encode them with fountain codes, so any frame that gets captured helps rebuild the file, not just a specific one in sequence.
Stream those blocks as a continuously changing QR code on one phone or laptop screen. No network connection needed at all.
Point a second phone's camera at the screen and let it collect frames as fast as it can, in any order, dropped frames and all.
Once it's gathered roughly 15% more frames than there are blocks, reconstruct the original file and verify it with a hash check.
Worth knowing: this exact idea already existed in a few other projects, and the creator credits them directly instead of pretending otherwise. Good execution still counts, even when the idea isn't brand new.
Here’s the GitHub Skill and this is the live app.
🎤 PAI Weekly Live Build
Every week, the Practicaly AI team hosts a live AI build session where you bring the ideas and workflows for your work or business and we help you build the automated workflow live.
Because we build in real-time, you get to see how we think through the problem, where we get stuck, and what the final outcome is. It's the difference between reading a recipe and watching someone cook it while talking through their decisions.
🗓️ When: Friday, August 7 2026 @ 10AM PT (1PM ET)
📹️ Where: Virtual
Did you find today's post useful?
Until next time,
Team @PracticalyAI
