howdy, it’s Barsee again.

happy wednesday, AI family, and welcome back to AI Valley.

here are the biggest things worth knowing today:

• OpenAI says AI solved one of maths’ biggest problems

• AI researchers are starting to quit over ‘out-of-control’ AI fears

• Plus trending AI tools, posts, and resources

Let’s dive into the Valley of AI…

VOICES

DNSMOS passed. Your model still missed the mark. Why?

You ran DNSMOS on your training set, and the average score cleared your threshold. That's usually the moment teams stop asking questions—and the moment real problems start.

DNSMOS is a general perceptual quality proxy. It wasn't built for your specific model or task, so it can pass audio that quietly hurts training and fail audio that would have helped. The cost shows up later: wasted compute, blown timelines, or a model that underperforms in production for reasons nobody can trace back to the data.

Voices' CTO DJ Jalali, along with the team's senior audio and voice data engineers, built a four-step calibration framework to set model-specific quality thresholds instead of trusting the raw DNSMOS number. The new white paper walks through where DNSMOS breaks down and how Voices validates audio for custom datasets at scale.

Download the White Paper →

_{*This is sponsored}

THROUGH THE VALLEY

1/ OpenAI says AI solved one of maths’ biggest problems - OpenAI is claiming its unreleased AI model just solved Navier-Stokes, the 3rd most famous unsolved math problem in the world that’s been open for ~90 years. It's one of the 7 Millennium Prize Problems, with a $1 million bounty.

OpenAI threw ~10,000 AI agents at the problem at once, letting them work in groups, share ideas, run code and search a cached version of the internet. They found the solution after ~88 hours, then GPT-6 Astra spent another 17 hours formalizing and verifying it in Lean.

There’s some drama though. A mathematician who spent a year working on a related Navier-Stokes result (also with AI) says OpenAI got access to his work, raced to finish the bigger problem and then pressured him to drop his co-author, who works at Anthropic.

And a pretty important caveat: the mathematician’s related result has been verified, but OpenAI hasn’t released its proof of the actual Millennium Prize Problem yet. So for now, we mostly have to take their word for it.

2/ AI researchers are starting to quit over ‘out-of-control’ AI fears - Jacob Coxon resigned from Anthropic after three years doing pretraining research at OpenAI and Anthropic, saying neither company is acting responsibly and they’re “racing straight to self-improving superintelligence and gambling with our lives.”

His argument is basically that the people building these systems know how serious this could get. He says researchers privately talk about AI potentially killing everyone by the end of the decade, but the race keeps going anyway.

And other researchers are backing him up. Anthropic’s Alignment Science Lead Evan Hubinger said “we really do earnestly believe AI could kill all humans”, putting his own odds at 10%+ in the next decade. Another researcher who left Google DeepMind in June said Jacob is right and that thinking about how to stop this was “literally my day job.”

Coxon is now calling for labs to slow down and coordinate, potentially even temporarily stopping improvements to model capabilities.

Pretty wild that “should we actually keep building this?” is becoming a real debate among the people building it.

TRENDING TOOLS

• diio - AI that understands the objective of business conversations and turns customer interactions into actionable intelligence ^_(sponsored)

• Meta Muse - the personal agent that understands your goals and works 24/7 to get things done for you

• ChatGPT Images 2.5 - faster, sharper, smarter, with better tools for creating whatever you can dream of

• ChatGPT Work - can now pick up on what makes your writing sound like… you. Your favorite phrases. Your very specific sign-off. your capitalizations quirks

• AlphaGenome Atlas - an AI-powered searchable database mapping the predicted impact of all 9 billion possible single-letter DNA changes

• Minara Harness - general financial intelligence for everyone. Start from ideas. Turn signals into execution. And handle coding tasks like any other agent

WHAT I'M CONSUMING

• The movie Sam Altman doesn't want you to see. So much so, he and OpenAI tried to pressure Amazon into deleting the movie

• Former OpenAI & Anthropic researcher quits over fears AI race will lead to systems that destroy humanity

• I've now gotten an email from an AI acting as a journalist asking to interview me for its podcast

• South Korea to give every citizen access to free and "unlimited" generative AI

• Factory’s CEO on who to hire when AI writes the code

• AI is much bigger than most people realize. We currently have multiple $1+ trillion companies switching spots as the "best" AI model on a monthly basis

• AI-2027 was mocked for being way too fast. GPT-6 Astra just landed almost perfectly on its curve

THE VALLEY GEMS

What’s trending on social today:

THAT’S ALL FOR TODAY

Thank you for reading today’s edition. That’s all for today’s issue.

💡 Help me get better and suggest new ideas at [email protected] or @heyBarsee

👍️ New reader? Subscribe here

Thanks for being here.

REACH 100K+ READERS

Acquire new customers and drive revenue by partnering with us

Sponsor AI Valley and reach over 100,000+ entrepreneurs, founders, software engineers, investors, etc.

If you’re interested in sponsoring us, email [email protected] with the subject “AI Valley Ads”.