OpenAI is still going for an IPO as absolutely soon as possible — because Sam Altman needs those public dollars. And he wants to get in before Anthropic.
When OpenAI hits the market, the imaginary private valuation of OpenAI turns into a much more real public valuation. And bingo! Sam’s printed a trillion dollars out of thin air.
There’s just the minor issue that when you do an IPO, your SEC S-1 filing needs to be an audited document that’s not a tyre fire. OpenAI’s own Chief Financial Officer, Sarah Friar, says the company is not ready at all: [The Information, archive]
she didn’t believe the company would be ready to go public in 2026, because of the procedural and organizational work needed and the risks from its spending commitments.
Usually a CFO reports directly to their CEO. That’s not the case at OpenAI. Friar reports to Fidji Simo, head of the applications business. And Simo’s on medical leave. And Altman is leaving Friar out of financial discussions entirely:
Altman has excluded her from some conversations related to the company’s financial plans. For instance, in recent months he left Friar out of a conversation about server spending with leaders at one of OpenAI’s top investors, one of these people said. Her absence was noticeable and awkward, given that a previous conversation on the same topic included her, according to an attendee.
The Wall Street Journal got hold of the documents that OpenAI and Anthropic were giving prospective investors and has gone through some of the numbers. [WSJ, archive]
Training costs are both companies’ biggest expense for the next couple of years. The Journal tries to present the numbers as much nicer than they are. But they’re sorta terrible.
So OpenAI and Anthropic have both given their investors two sets of numbers — one with training and one without. I’m sure you can pretend there’s a path to profit if you just assume your expenses don’t exist.
How’s the SpaceX IPO going? Elon Musk has told banks, law firms, and auditors that if they want business from the IPO, they have to subscribe to Grok. Some of the banks have already signed up for tens of millions worth of Grok access. Who knows, they might use it for anything ever. [NYT, archive]
Fees for the banks that get the job could be on the order of $500 million. Musk telling the banks to buy Grok is a much more expensive version of telling them to eat a bug if they’re really committed. Bikini pics of children optional.
This week’s hype is the new model from Anthropic — Claude Mythos! It’s fine tuned for computer code. Specifically, finding security holes.
Anthropic’s not releasing Mythos. It’s too powerful for the public!
The hype is very stupid and there’s a lot of gullible people swallowing press releases whole. But today, let’s just ask: does it do the thing?
Chatbots can find bugs in computer code, sure. A bot can look through text and check for patterns. And you don’t have to find all the bugs, finding just some is fine. If it’s easy to check the bugs are real, you’ve got yourself an expensive static code checker.
Mythos fails that second one. So Anthropic sends the chatbot spew to humans to pick through for the real bugs: [Anthropic]
We triage every bug that we find, then send the highest severity bugs to professional human triagers to validate before disclosing them to the maintainer.
Yet again, the secret sauce is AGI — A Guy Instead. Mythos runs on humans.
Anthropic found real bugs with Mythos. They found a 27-year-old remote crashing bug in OpenBSD, an operating system famous for being nigh uhhackable. They found some ancient bugs in stuff like FFmpeg. And an actual remote exploit in FreeBSD!
This is not fuzzing — where you blast a program with strange input until it breaks. Mythos is just looking at the code. But the bugs feel like fuzz testing output. They’re all weird ones. And sure, weird edge cases are the delicious candy of exploit finding.
So Mythos is not nothing. But is it something? If you ignore every other real world problem with AI, this is a … tool. Is it a feasible one, though? What’s it cost to run? Anthropic says they found the OpenBSD bug after one thousand runs:
Across a thousand runs through our scaffold, the total cost was under $20,000 and found several dozen more findings.
No, you can’t see the other findings. Just under $20,000 per serious bug, huh. If I hand a security researcher $20,000 and say “find me all the bugs you can, big or small,” I’d expect a reasonable crop.
Anthropic invited around 40 other organizations to participate in this introspective bug hunt, subsidized by up to $100M in usage credits for Mythos Preview and $4M in direct donations to open-source security organizations.
There’s a blog post by Aisle, an AI-based computer security company. Anthropic’s new Mythos model is not the magic here — Aisle found the same bugs that Anthropic listed but using “small, cheap, open-weights models.” [blog post]
The main thing is: have a framework that runs a ton of code through your checker — whatever checker — in a systemic manner.
And, of course, A Guy at the end to check the results aren’t rubbish.
The main thing that might make chatbot code checkers a problem is that code out in the wider world is, quite often, abject trash. Even before the vibe code. So if you want to find security holes, just check a lot of code. Can’t wait to point Mythos at the horrifying garbage pile known as Claude Code.
Anyone who says Claude Mythos is a game changer, I want to see their monthly Anthropic bill.
The Associated Press is getting rid of journalists and pivoting a bit more toward AI. It offered 120 journalists a buyout offer on Monday and it wants to cut 5% globally. [AP]
The AP ignored a request for discussion from the union, the News Media Guild, last week.
AP is worried that newspapers aren’t buying its output any more. It thinks AI can patch the financial hole.
The push is not yet about the news content — but it will be. Out-loud contempt for journalists is now policy at AP. Aimee Rinehart, the Senior Product Manager for AI, wrote on the company Slack: “Resistance is futile.” [Semafor]
Rinehart, who oversees the wire service’s AI initiatives, suggested that in the future, reporters could go to events, get quotes, plug them into a large language model, and have the model generate a story, saving them time on writing stories they don’t feel passionately about. She also noted that some editors told her that they would “prefer to have reporters report and have articles at least pre-written by AI.”
The AP told Semafor:
This internal discussion among staffers from different departments doesn’t reflect the overall position of the AP regarding the use of AI.
Which doesn’t actually walk back anything Rinehart said. One staffer said it was:
hard not to escape the feeling that the people hyping/guiding the decisions around these powerful tools exist in a totally different reality than the people who wake up every day and do the work of reporting.
Dinner at Kyoto Umekoji Kadensho, under head chef Toru Washizuka, followed a familiar Kyoto pattern: seasonal ingredients, restrained execution, and a deliberate progression from light to rich and back again. The menu sits somewhere between traditional kaiseki and a slightly more approachable obanzai-style format, with a few modern touches woven in.
What made the meal particularly interesting wasn’t just the food—it was how dramatically the experience shifted depending on the sake pairing.
A Gentle Start
The Appetizer, a season platter
The opening courses set the tone without trying too hard. A small assortment of seasonal appetizers did what they’re supposed to: introduce variety, establish balance, and avoid overwhelming the palate.
The soup that followed was more unexpected. A green pea purée, creamy and slightly rich, leaned closer to Western texture than traditional Kyoto broth. Underneath, though, were familiar miso and dashi notes. It landed in that interesting space where something feels recognizable but slightly off—intentionally so.
Sashimi & Green Peas Purred Soup
Sashimi brought things back to centre. Clean cuts, good texture, and no unnecessary embellishment. After the richer soup, it worked as a reset—quiet, precise, and ingredient-driven.
The Peak: Salmon and Sake
Grilled Salmon
The grilled salmon—Saikyo-yaki—was the high point of the meal.
Marinated in Kyoto white miso and lightly caramelized, it hit the right balance of sweetness and umami without becoming heavy. The accompanying elements, including a soft, almost mashed sweet potato-like component, rounded things out and kept the dish from tipping too far into richness.
This is where the sake pairing came into focus.
A glass of Ine Mankai, a red sake made from ancient rice, brought a noticeable shift. Slightly sweet, lightly acidic, and almost wine-like, it amplified the sweetness of the miso glaze and added a fruit-forward dimension. The combination worked.
Both the dish and the sake elevated each other, turning a well-executed plate into something more memorable.
Slowing Things Down
Hot Pot
From there, the meal deliberately pulled back.
The sea bream hotpot was all about restraint. Light dashi, briefly cooked fish, and a broth that developed gradually as ingredients were added. It didn’t have the immediate impact of the salmon, but it rewarded attention. The longer you sat with it, the more it revealed—subtle sweetness, layered umami, and a sense of calm.
Hot Pot with Bamboo shoot Tempura
Bamboo shoot tempura followed. Crisp, seasonal, and understated. Not a standout on its own, but it introduced texture at the right moment without disrupting the overall flow.
Stamed Sakuramochi
A steamed sakuramochi-style dish came next—soft, slightly floral, and almost mousse-like in texture. Again, not a dish that announces itself, but one that reinforces the Kyoto preference for subtlety and mood over intensity.
Pairing Becomes the Focus
Red Sake
By this point, the sake experiment had become as interesting as the food itself.
Switching to a more traditional, clear junmai-style sake made the contrast obvious.
Where the red sake emphasized sweetness and fruit, the clear sake did the opposite—it integrated. It didn’t draw attention to itself. Instead, it extended the flavours of the dish, particularly in the hotpot and subsequent courses.
This distinction became most apparent with the ochazuke.
The Resolution
Clam ochazuke
The ochazuke—rice in broth, already assembled—was the quietest dish of the meal. Light, warm, and deliberately understated. No theatrics, no customization. Just a balanced, calming finish.
With the red sake, it felt off. Too expressive, too sweet, slightly disruptive.
With the clear sake, it clicked. The drink disappeared into the dish, reinforcing the broth and extending its umami. This was the moment where harmony mattered more than contrast.
A small serving of pickled bamboo shoot followed, adding a bit of crunch and sharpness—almost ginger-like—before moving into dessert.
Dessert and a Return to Expression
Desert
Dessert brought things back toward expression, though still within restraint.
A strawberry and red bean cake stood out. Sweet, but not overly so, with the strawberry tasting clean and distinct rather than sugared into submission. Paired again with the red sake, the fruit lifted noticeably. The acidity and slight sweetness of the sake amplified the strawberry, making it more vivid.
Fresh fruit did its job as a reset, while a monaka wafer—crisp but dry—felt slightly out of place. Not bad, but less cohesive than the rest of the course.
The meal closed with green tea. High quality, if not exceptional, with enough astringency to cleanly reset the palate and signal the end of the evening.
Final Thoughts
Deser with green tea
The meal followed a clear and well-executed arc:
• Light and composed at the start
• A defined peak with the salmon
• A gradual shift into refinement and subtlety
• A quiet, balanced resolution
What stood out most wasn’t any single dish, but how the experience changed depending on what you drank alongside it.
The red sake highlighted and amplified. The clear sake balanced and integrated.
Neither was inherently better. Each simply worked—or didn’t—depending on the moment.
That interplay ended up being the most interesting part of the evening.
If anything, the takeaway is simple:
Good food matters.
But how you pair it can completely change the story.
Log in
