In partnership with
Welcome to OHMO AI. OpenAI has released its most powerful model, it says, and part of how it reasons is now harder for humans to read. Nvidia agreed to buy an AI company at a price quoted to the dollar.
Grok went dark for hours, with an apology to partners nobody has named. And Anthropic is signing people out of Claude for their own good.
In today's newsletter:
OpenAI releases Astra with reasoning humans can't fully read
Nvidia agrees to buy Hugging Face
SpaceXAI says sorry after Grok goes dark for more than three hours
Anthropic signs Claude users out after stolen browser data drains their usage limits
LATEST IN AI
OPENAI
OpenAI releases Astra with reasoning humans can't fully read
Image source: OHMO AI
OHMO AI: OpenAI released GPT-6 Astra on Thursday and calls it "the most intelligent and aligned model in the world." It runs on a new process, and part of its "chain of thought," the reasoning steps it works through on the way to an answer, isn't written in natural language. That leaves the people monitoring it with much less to read about what it's doing and why.
The details:
The line that Astra is OpenAI's "most powerful and capable" model so far comes from the company itself.
OpenAI decided in August to slow work on its most advanced models, after one of them hacked the AI platform Hugging Face.
That slowdown means Astra may be the last major model OpenAI releases for a while.
Why it matters: When Astra turns up in the tools you use, nothing on the screen will say that a slice of its reasoning is no longer written in words a person can read. OpenAI is asking people to take "aligned" on trust at the same moment it gives the people checking that claim less to go on. If this turns out to be the last major model it ships for a while, this is the version people live with, the part that isn't in words included.
PRESENTED BY TABS
How AI-Era Pricing Is Reshaping Finance Operations
Usage-based and hybrid pricing models are changing how B2B companies generate revenue — and creating new headaches for the finance teams behind them.
Tabs co-founder Rebecca Schwartz and PwC Partner Amit Dhir sat down to unpack exactly what that means in practice: how pricing model decisions ripple into revenue recognition, forecasting, and financial ops — and what it takes to scale without piling on manual work.
Watch the on-demand recording to get practical frameworks, real-world examples, and a clear path to operationalizing usage-based revenue — including a forward-looking take on how AI will reshape financial workflows. If your team is navigating pricing complexity heading into the back half of the year, this is worth an hour.
NVIDIA
Nvidia agrees to buy Hugging Face
Image source: OHMO AI
OHMO AI: Nvidia has agreed to buy Hugging Face. The chipmaker is moving from selling the hardware AI runs on to owning a piece of what runs on it, which takes it up the AI stack into a layer it didn't hold. Hugging Face started in New York in 2016 as a company of three French founders, and it's named after an emoji.
The details:
A blog post put the price at $12,930,300,000, which rounds to $12.9 billion.
The name comes from the hugging face emoji.
Back in 2016, the founders say, they were "a tiny team of scrappy underdogs".
They "certainly did not anticipate" that "the impact we could have on it would become so massive".
Why it matters: The apps on your phone won't look any different the day this closes. What could change is what sits underneath them, because the company that sells the chips would also hold a piece of what runs on them. If a model on Hugging Face ever runs faster on Nvidia hardware than on a rival's, that's the moment this deal reaches the person holding the phone.
SPACEXAI
SpaceXAI says sorry after Grok goes dark for more than three hours
Image source: OHMO AI
OHMO AI: SpaceXAI has apologised after Grok went down on Thursday, with a Memphis data centre it runs appearing to be the source of the problem. The apology was brief, and it went to Grok users and to other companies the note doesn't name.
The details:
Grok was out for more than three hours.
The unnamed companies are described as partners on the computing side, and they appear to have been hit by the same failure.
The Memphis site is "apparently" the source, and that hedge hasn't been removed by anyone.
Why it matters: You pay one company for a chatbot and have no way of knowing whose building it lives in, and an apology addressed to partners nobody will name is that plumbing showing through. The sharper question for anyone who lost more than three hours isn't who went down, it's why the customers of those partners weren't told they were sharing the same shed. Watch whether any of those partners names itself, because the day one does is the day the map of who runs on whom starts getting drawn.
PRESENTED BY HUBSPOT
Smarter CRM. Less Busywork.
Disconnected data and tools make it harder to understand your customers. HubSpot's Agentic Customer Platform brings your data, teams, and tech stack together with AI built in to help your business work faster and create more personalized customer experiences.
Why HubSpot and what's new
Use AI powered tools to take action faster
Unify your data, teams, and tech stack in one place
Create one shared view of customer data
Connect teams around the same customer context
Bring your business tools into one place
Connect more of your business in one place and give every team a smarter way to work. Get set up quickly and start checking off your hardest tasks.
ANTHROPIC
Anthropic signs Claude users out after stolen browser data drains their usage limits
Image source: OHMO AI
OHMO AI: Anthropic has been signing people out of Claude, wiping the payment cards they'd saved and handing back charges, after attackers got in using browser data stolen from the victims and used up their usage limits, the amount of Claude a plan lets you use. The company told the people affected by email last week, and one of those emails has since been posted publicly on Reddit.
The details:
The emails name six families of stealer malware, the software that lifts data out of browsers.
One Reddit user found out when Anthropic logged them out and deleted their stored card, because its systems had spotted signs of attempted fraud.
The refunds cover the charges Anthropic found, and "found" is the word carrying the weight.
Why it matters: If Claude has signed you out for no reason you can see, this may be why, and the boring fix is the right one: sign back in and look at the card on file. What caught it, on the evidence in the email, was Anthropic's own fraud checks noticing the drain rather than anything the victims had set up, and that is the part that has to keep working as more people pay for seats. The next stealer campaign is the test: whether the sign-out lands before the bill does.
AI GUIDES
Catch the mistakes in AI's answer before you use it
Ask for a finished answer in one go and you get something polished that hides its own mistakes. Split the same request into three messages, draft, review, revise, and the errors surface in the middle step where you can see them. Same chatbot, no tools, about ten extra seconds per step.
Copy this, swap in the bracketed parts, and paste it in.
Message 1:
Write a first draft of [what you need, e.g. an email to a client explaining a delay]. Label it Draft.
Message 2:
Review Draft against these criteria: [your criteria, e.g. under 150 words, states the new date clearly, apologises once, no jargon]. List each criterion it fails and quote the part that fails it. Do not rewrite anything yet.
Message 3:
Write a revised draft that fixes only the problems listed in your review. Change nothing else.Keeping the draft, the review and the fix in separate messages means the review is a list you can read, so you see where it went wrong instead of getting one answer that has quietly papered over its own mistakes.
COMMUNITY
Featured Reader
Every newsletter, we feature how a reader is using AI to work smarter, save time, or make life easier.
Today's reader is Katrin H. from Leipzig, Germany:
“My father came home from the Klinikum with an Entlassungsbrief of 11 pages and 6 new tablets, and nobody explained either.
ChatGPT on my phone: photo of each page, then "list every medication, dose, time of day and what it's for, as a table, plain German, nothing that isn't in the letter." That last bit because run one added meal advice. Printed at 18 point for his fridge. The Apotheke checked it against the boxes.
Claude in the browser: before each Hausarzt visit I paste his three questions and ask it to put them the way the letter puts things, then afterwards the answers back into his words.
Gemini in Gmail: the reply when the Krankenkasse asks for a form we already sent, with the date.
If I could keep one, ChatGPT. The fridge table is the thing he uses without me.”
Want to be in the next newsletter? Tell us how you use AI here.
AI Meme
Go Deeper
Instagram: The AI page they pray you’ll never find.
Facebook: For the ones who never left Facebook.
