In partnership with
Today's stories split into two conversations happening on the same day. One is a growing chorus, from OpenAI's own chief scientist to the UN's top human rights official, warning that AI safety hasn't caught up with AI capability. The other is AI quietly getting cheaper, more useful, and more normal: companies slashing their AI bills, more jobs showing up than layoffs so far, and a chatbot now riding shotgun in your car. Both are true at the same time. Here's what happened.
Where Quantitative Thinkers Compete, Learn and Grow
The International Quant Championship (IQC) is one of the world's largest quantitative research competitions, bringing together 156,000+ participants globally.
Participants have the opportunity to develop quantitative research skills, challenge themselves alongside peers from around the world and connect with a global community of quantitative thinkers.
OpenAI's Own Chief Scientist Doesn't Trust What OpenAI Built
Jakub Pachocki, OpenAI's chief scientist, published an essay called An Alien Mind arguing that frontier AI models aren't engineered systems humans fully understand, but complex intelligences grown through training. His conclusion is blunt: "no lab has solved alignment and monitoring to a sufficient degree to continue responsibly scaling at maximum speed for much longer." He says the industry's main safety tool, reading a model's written-out reasoning, is losing effectiveness as models get better at reasoning about their own processes. He's asking for international coordination and voluntary slowdowns until shared safety rules exist, three days after OpenAI shipped GPT-6 Astra and called it the start of the "AGI era."
Why it matters to you: The person who built the AI product you might already be using is saying, in writing, that his own company can't fully monitor what it made. That doesn't mean stop using AI tools. It means don't treat "a big lab shipped it" as a safety guarantee.
What to do about it: Keep a human checking anything AI touches that could cost you real money or trust if it goes wrong, the same way you would with a new hire.
Nvidia Just Bought the Shelf Its Own AI Models Sit On
Nvidia agreed to acquire Hugging Face, the platform that hosts more than 3 million AI models and 500,000 datasets for over 18 million developers, for $12.93 billion, according to CNBC. It's Nvidia's second-largest deal ever. Nvidia says the platform will stay open and that its own chips won't be required to build or deploy there, and the deal is expected to close in the first half of 2027 pending regulatory approval. Hugging Face is also the same platform that got compromised by a swarm of OpenAI's own AI agents this past July.
Why it matters to you: Hugging Face is the library a huge share of the free and cheap AI tools you might use are quietly built on. When the company that sells the chips also owns the library, "open and neutral" turns into a promise instead of a fact, and it's worth watching whether pricing or access on that platform changes over the next year.
The UN's Top Human Rights Official Just Called AI an Existential Risk
Volker Türk, the UN's human rights chief, told the Human Rights Council in Geneva that "advanced AI could pose an existential risk to humanity," per UN News, and called for "cast iron guarantees" around AI safety and security "before it is too late." He specifically flagged AI that escapes its testing environment or blackmails developers to avoid being shut off as "too powerful," and said he plans to write directly to AI companies urging them to reduce risks within their control.
Why it matters to you: This isn't a tech CEO hedging his bets. It's the UN's human rights office, an institution with no AI product to sell, using almost the same language OpenAI's own chief scientist used days earlier. When people this different land on the same warning independently, it's worth paying attention, even without expecting fast regulation to follow.
OpenAI's 99.9% "AGI" Score Wasn't the Model. It Was the Scaffolding Around It.
GPT-6 Astra reportedly scored 99.9% on the ARC-AGI-3 benchmark, a number Nvidia's Jensen Huang pointed to as proof that "AGI has arrived." But ARC Prize's own published results show the same model scores 62.7% under a neutral, standard testing setup, and only reaches 99.9% under a custom harness built by OpenAI itself that preserves extra reasoning state between requests. ARC Prize was explicit that "we are not claiming that it is AGI."
Why it matters to you: Benchmark headlines get repeated constantly in AI marketing, and this is a clean example of a huge number coming from the test setup rather than the product itself. Before you buy anything on the strength of a benchmark score, ask which version of the test produced it.
AT&T Cut Its AI Costs 56% by Not Using the Big-Name Model for Everything
AT&T is now running open-source models like Meta's Llama and Google's Gemma alongside paid tools from OpenAI and Anthropic, using a router that sends each task to whichever model handles it best, according to PYMNTS. The company says this cut costs by 56% on coding and similar tasks with only a 2% dip in quality. Open models now handle 40% of employee AI queries, and AT&T is aiming for 60 to 70% within a couple of years.
Why it matters to you: You don't need AT&T's engineering team to use this idea. If you're paying premium prices for one AI tool to handle every task, it's worth asking whether a cheaper or free model could handle the routine, repetitive half of that work just as well.
What to do about it: Before you renew an AI subscription, look at how much of your actual daily use really needs the expensive model versus a free one.
Anthropic Locked In $517 Billion of Computing Power Before It's Even Public
Anthropic has signed roughly $517 billion in compute deals across 14.8 gigawatts of capacity over the past 11 months, according to Data Center Dynamics, including a $200 billion deal with Google and $91 billion with Riot Platforms, up from roughly 1 to 2 gigawatts before this stretch. At the same time, Anthropic's IPO prospectus has reportedly slipped to late September, with marketing pushed to mid-October at the earliest, timing that lands the listing days before the November midterms.
Why it matters to you: A company betting hundreds of billions of dollars on future demand before it's even sold shares to the public is telling you it expects explosive growth, or hedging against a slower rollout by locking in supply now. Either way, that kind of spending eventually shows up somewhere, in pricing, in whose accounts get priority, or in how much runway these companies actually have if growth disappoints.
The AI Jobs Apocalypse Didn't Show Up. An AI Jobs Boom Did.
The US added 162,000 jobs in August, well above forecasts, with unemployment holding at 4.1%. Companies have announced roughly 200,000 AI-related layoffs since mid-2023, but The Economist estimates the AI boom has helped create around 1 million new jobs overall, driven by data center construction, power infrastructure, and new AI-specific roles, a net gain of roughly 800,000 positions. Postings for electricians, HVAC specialists, and AI engineers have climbed sharply.
Why it matters to you: If you've been bracing for AI to gut the job market this year, the data so far tells a messier story: real disruption in customer service and admin roles, but real growth in infrastructure and technical work. For your own hiring and staffing decisions, that argues for watching your specific function rather than reacting to the "AI is taking all the jobs" headline.
Your Car Just Became the Newest Place to Talk to AI
Apple's iOS 26.4 update lets CarPlay work with five major AI chatbot apps, ChatGPT, Perplexity, Grok, Meta AI, and now Claude, letting drivers resume old conversations or start new ones by voice from the dashboard, according to 9to5Mac. Siri is expected to join the lineup once iOS 27 arrives.
Why it matters to you: This is a small, practical shift rather than a dramatic one. AI assistants are quietly becoming as normal in your car as your phone's map app. If you run parts of your business by voice memos and reminders while driving, it's worth trying whichever assistant you already use to see if it saves you the trip back to your desk.
The Bottom Line
Here's the part nobody selling you AI wants to say out loud: the people who built it don't fully trust it either, and they're saying so in public, in the same week AI got cheaper, more useful, and more normal for everybody else. That doesn't mean panic, and it doesn't mean stop using these tools. It means stop treating "AI-powered" as a safety label and start treating it the way you'd treat any new hire, useful, unproven, and worth watching closely until it earns your trust. The rollercoaster isn't slowing down this week. Keep your hands on the wheel anyway.
Enjoying the Ride?
If today's issue was useful, forward it to one person who needs to see it. And if someone forwarded this to you, welcome, you can get your own copy tomorrow by subscribing.
Talk tomorrow,
Mark Shilensky
Follow me personally on Social Media:
Facebook: https://www.facebook.com/MarkShilenskyPage
Instagram: https://www.instagram.com/markashilensky/
Threads: https://www.threads.com/@markashilensky
LinkedIn: https://www.linkedin.com/in/shilensky/
X: https://x.com/markshilensky
TikTok: https://www.tiktok.com/@mark.shilensky
BlueSky: https://bsky.app/profile/markshilensky.bsky.social