Welcome, humans.

So apparently the Universal Music Group music label just teamed up with AI voice model leader ElevenLabs to create a licensed AI remix platform. 

It’s still in development, but artists can opt in, and fans will be able to remix, mash up, and reinterpret their songs; y’know, respecting artist's’ copyright, but bringing everybody (including normal people) into the remix industry for funsies.

Meanwhile, Suno’s v6 family is pushing the tool side. Suno says the new fam is 5× faster than v5.5, with higher fidelity and fewer artifacts. It can edit one section in plain English, mash songs together, sample a riff, or swap one lyric. Try it here.

Corey made a good point about UMG remix platform: TBH, the music industry hasn’t really been analog for a long time. What you hear is already covered in digital production, effects, edits, samples, pitch correction, and other invisible layers.

If this ends with artists getting paid when people remix their work, and a regular person with one killer idea can get paid too, that seems like a net positive. The remix was coming either way. Paying both people seems better than pretending the button doesn’t exist.

Here’s what happened in AI today:

• 😺 OpenAI made progress on another Millennium problem.

• 📰 Anthropic detailed five biological-misuse cases.

• 📰 OpenAI launched ChatGPT for Financial Services.

• 🍪 OpenAI put Codex’s harness behind an Agents API.

• 🎓 Build AI “evals” around your actual job.

…and a whole lot more that you can read about here.

😺 OpenAI says it made progress on ANOTHER Millennium Prize problem

OpenAI’s whole 10,000-agents solve a $1M math problem thingy may already have a sequel.

In a statement reported by The New York Times, OpenAI said it had made “substantial progress” on another Millennium Prize problem and was working out how to share the result. It did not, however, name the problem.

Andrew Curran says the rumor is the Hodge Conjecture and speculates the unreleased model may be called Aeon. OpenAI has confirmed neither detail.

Here's what happened:

• OpenAI says an internal model significantly more capable than GPT-6 Astra produced its proposed Navier-Stokes solution and a formal proof in Lean.

• The successful run coordinated roughly 10,000 agents for about 88 hours, letting groups explore and share mathematical approaches in parallel.

• OpenAI now says it has made substantial progress on a second Millennium Prize problem since that run.

• Curran says the Hodge Conjecture is the leading rumor, but that remains unconfirmed.

Okay, so how can throwing more compute at old math problems suddenly move them?

The useful trick is checkability. Thousands of agents can explore different proof ideas and share promising paths, while code and formal verification make proposed answers easier to test. That turns extra compute into more shots on a checkable result. And suddenly, you can throw 10,000 PhDs at anything…

Our take: The Hodge rumor matters less than whether the result survives independent review. Watch for a full proof, formal verification, and outside mathematicians weighing in and give it the A-OKAY (The Terrence Tao +1, if you will). If OpenAI can repeat this across unrelated problems, verified research output becomes a stronger capability signal than another leaderboard score.

FROM OUR PARTNERS

Your AI problem probably isn’t “should employees use it?” They already are, and IT may not see half of it. On September 15, Tines VP of Product Colin Bentley will show you how to find the “wild code” already running across your business before it turns into risk or technical debt.

Learn how to:

• Spot the AI-powered code already running across your business

• Give teams freedom to build without piling on risk or technical debt

• Turn AI speed into something you can actually govern

Save your spot

🎓 AI Skill of the Day: Turn your AI corrections into a personal benchmark

Public benchmarks can tell you how models perform on generic tests. Every’s new workflow turns the corrections you already make into a personal benchmark for your actual job.

1. Pick one recurring task and save the original prompt plus source files.

2. Review the first output and turn each correction into one yes-or-no check, like “one idea per slide.”

3. Grade the output yourself, then have AI grade the same checklist without seeing your scores. Tighten any check where you disagree, then rerun the same test across models.

Every found GPT-5.6 Luna beat larger models on many of Mike Taylor’s everyday tasks. That is the point: measure the cleanup you actually care about.

They’ve even got prompts for you to copy in the link above… definitely check it out!

Have a specific skill you want to learn? Request it here.

FROM OUR PARTNERS

Build with AI. Skip the gatekeeping.

Edge Case is Akamai Cloud’s open Discord for developers and architects building with AI/LLMs, serverless, cloud infrastructure, and WebAssembly. Join live builds, grab production-ready patterns, and claim $300 in credits. Join today.

📰 Around the Horn

1. Anthropic said it disrupted Claude misuse across cyber, surveillance, and five biological cases that could support weapons development, while stressing it did not assert the scientists intended harm

2. OpenAI launched ChatGPT for Financial Services with GPT-6 Astra, built-in premium datasets, citations, and workflows for modeling, research, and pitchbooks.

3. Codex lead Tibo says OpenAI will pause new $200 Pro signups to protect Astra access for existing users under heavy demand; other plans and the API remain open.

4. Semafor reported Andrew Tulloch is leaving Meta after delaying his departure until the company launched its new open-source model family and Muse assistant.

5. NVIDIA partnered with Australian data-center providers on capacity for up to 2 gigawatts of AI infrastructure by 2027.

6. Visa, Mastercard, and Ant International launched a shared effort to identify and verify AI agents that make purchases for users.

7. A year-long Character.AI study found heavier companion use was associated with lower well-being and less human social interaction over time, especially among people using the bots for companionship or self-disclosure.

🍪 Treats to Try

*Asterisk = from our partners (only the first one!). Advertise to 700K+ readers here!

1. *Discover the potential of artificial intelligence with our comprehensive cheat sheet. Learn more about the concepts, platforms and applications of AI.

2. MultiMatte removes everything except the object you name, so “keep the jeans” can become a clean product cutout.

3. Genspark launched Gen-1 Slides, its first proprietary model, built on MiniMax with Fireworks AI; Genspark says its slide quality matches frontier models at roughly 1/17th of Claude Opus 5’s price.

4. Ryan Carr of Moodboard gives you a Lead Magnet Wizard prompt for Astra that interviews you, pitches calculator/scorecard/quiz ideas around one customer problem, then helps build the one you pick.

5. Google Labs’ Dreambeans turns selected Gmail, Calendar, Photos, Search, YouTube, and Gemini context into finite daily reminders and recommendations, now for U.S. users 18+ on Android and iOS.

6. Speak’s Live Tutor Lessons puts GPT-Live-1 into limited English and Spanish practice; SpeakBench says it cut interruptions during model “thinking pauses” by about 80%.

Developer stuff

1. OpenAI put GPT-Live-1 in the API, giving developers access to the $0.05/min full-duplex voice model that powers ChatGPT Voice; it can listen while speaking, handle interruptions and background noise, match tone and pace, and hand reasoning or tool use to another model behind the scenes.

2. OpenAI Agents API lets developers run long-lived cloud agents with Codex’s harness, tool use, context compaction, and subagents through one API.

3. Gemini for Windows puts an Alt+Space assistant beside your apps and can hand multi-step jobs to Gemini Spark.

4. DeepSeek V4.1 Flash gives you a native-vision, 1M-context open-weight model built to cut the cache burden of long agent workloads.

5. Cognition launched SWE-2, its Kimi K3-based coding model, then Devin Voice, which pairs SWE-2 with GPT-Live so you can talk through a coding job and hand it to Devin to build. Voice docs.

6. Cohere released North Small Translate, an open-weights 218B/25B-active MoE for 50+ languages that it says scored 83.5 on WMT26, above DeepL NextGen’s 81.2.

💡 Intelligent Insights

• Melanie Mitchell argues “rogue agent” and “escape” language can turn engineering failures into sci-fi metaphors, steering attention away from better sandboxes, independent tests, and liability.

• InventBuild.Studio argues AI makes competent copying cheap, so genuine creativity and editorial taste become the true scarce advantage.

• Roberto González at Brown argues Silicon Valley is now embedded deeper in the military-industrial complex, with about $28B in 2018–22 awards to Microsoft, Amazon, and Alphabet.

• Graybeard argues software teams go a little insane when changes are cheap and “done” has no physical stopping point; thankfully, regular customer contact dampens the effect.

• Julie Yoo at a16z argues AI could push employer-sponsored health plans into a replacement cycle by lowering administrative fixed costs as premiums rise more than 10% annually.

• Nathan Lambert argues recursive self-improvement will be lumpy, not a smooth takeoff: humans still bottleneck task creation and evaluation, while the final few percent of model quality can consume most of the work.

• Neel Nanda worries Astra can do more serial reasoning without an exposed chain-of-thought, which weakens safety methods that depend on monitoring the model’s visible reasoning.

• Ben Moll and Alex Imas argue exploding AI capability still probably will not produce double-digit U.S. GDP growth, because deployment runs into consumer, capital, cyber, and research bottlenecks.

NEW: WTF is GitHub? Watch the Replay of our Stream on GitHub For TOTAL Beginners!

Always wanted to make something with AI, but had no idea what to do after the chat spits out the code? Watch Cassidy Williams teach us how the from absolute basics, or read along with our total beginner’s guide as a handy cheat sheet, then dive deeper with GitHub’s beginner playlist.

Learn how to bring your digital ideas to life. No coding background required!

A Cat’s Commentary

Btw: We just launched a robotics newsletter! Sign up for it here.

P.S: Love the newsletter, but only want to get it once per week? Don’t unsubscribe—update your preferences here.