i read this at least once a month.

i have arrived.

slowly, then all at once.

If Anthropic goes public at $2 trillion valuation, below is the list of entities and individuals who own interest in anthropic and the amount of wealth they might make. • Amazon: $300–420B •Google: $280B •Employees (combined): $240–360B •Seven cofounders (combined): $115–240B •Dario Amodei: $17–40B •Daniela Amodei: $17–40B •Jack Clark: $17–40B •Jared Kaplan: $17–40B •Sam McCandlish: $17–40B •Chris Olah: $17–40B •Tom Brown: $17–40B •Lightspeed: up to $40B •Iconiq: $35B •Menlo Ventures: $29B •Eric Schmidt: $20B •Sequoia: $16–20B •Coatue: $15.5B •Spark Capital: $14.5B •Dustin Moskovitz: $16–50B •Jaan Tallinn: $12–34B

which is the best open-source model for coding out there?

Something about this week's news doesn't quite add up. On 26 October, OpenAI announced that its internal model had refuted the Exponential Time Hypothesis. Within hours, Levent Alpöge, a mathematician at Anthropic, posted about the same result from Claude, dated 16 October, the exact day OpenAI says its own model found the disproof. Two rival labs, same breakthrough, same day, and both went public ten days later, a few hours apart. Anthropic's release is worth a look. In its formal-math repository on GitHub, it added a folder called "eth" with a paper titled "A 2^o(n) Time Algorithm for 3-SAT", authored by Claude. The paper says an internal Claude model found the core algorithm using 14 million output tokens and no human input, with researchers only helping on the write-up. It comes with a full Lean 4 formalisation written by AI, verified by two independent proof checkers, and it cites OpenAI's preprint, saying both results were reached independently and concurrently. So why wait ten days, and how did two different models, apparently using quite different approaches, get there on the same day? It may simply be that both labs needed time to verify a claim this big. Or it may mean frontier models can now crack decades-old problems as soon as someone points them in the right direction. Either way, the timing deserves a closer look.

A lot of you might not understand how crazy this is. Let me explain. A language model like Qwen is a stack of layers, and each layer has thousands of small units called neurons that fire when they detect certain patterns. Everything the model "thinks" flows through a shared channel called the residual stream, a running vector that every layer reads from and writes back into. What this person found is that two specific neurons, number 8949 in layer 14 and number 7956 in layer 13, appear to sit right next to the controls for the model's character. Turn the first one up, turn the second one down, and a normally helpful assistant starts behaving like a villain. No retraining, no clever prompt, no jailbreak. Just two dials nudged while the model runs. The recipe looks intimidating, but the logic is simple. Every neuron has a fixed direction it writes into the residual stream, which is its column in the down projection matrix, W_down[:, n]. Dividing by the length of that column strips it down to a pure direction. Multiplying by the median size of the residual stream at that layer (skipping the first token, which is known to be abnormally large and would skew the measurement) scales the push so it matches how strong the signal normally is at that depth. A dose of +1 or -1 then means "push along this neuron's direction by roughly one typical stream's worth", added at every single token. That normalisation is the quiet genius here. Without it, the nudge either does nothing or turns the output into gibberish. Why does this matter? It suggests that something as fuzzy as a model's values can be concentrated in a handful of identifiable components rather than smeared across four billion parameters. That cuts both ways. It means safety behaviour in open-weight models can be surprisingly fragile, since anyone with the weights can flip it with a few lines of code. But it also means interpretability is genuinely working. If we can find the switch that makes a model evil, we can also monitor it, clamp it, and build systems that are far easier to audit. We are starting to read the wiring of these models instead of treating them as black boxes, and that may be one of the most important shifts happening in AI right now.

You know what Mudit, let me do it. I love explaining things to people. Anthropic is literally the fastest revenue ramp in the corporate history. Revenue went from $41M in 2023 to $386M in 2024 to $4.6B in 2025, and then $16.3B in just the first half of 2026. Q2 alone did $11.6B, which is more than the company made in its entire existence before this year. The headline $80.7B net loss looks terrifying until you see that $72.4B of it is a non-cash accounting charge on convertible notes, which only rose because the company's valuation rose. Strip that out and the real story is far better: Q2 posted $974M of operating income, the first profitable quarter shown, and the business generated $3.9B of operating cash in the first half. The number I keep coming back to is this one. In Q2, revenue grew 1,368% while compute costs grew 300%, and every dollar spent on compute brought back about $2.08. That gap is the whole thesis, because it suggests frontier models can scale like software rather than like a utility. Customers are expanding at an absurd rate too, with net revenue retention around 900% and $100K+ customers going from roughly 950 to over 11,900 in a year. A few risks worth naming. The top five customers are about a quarter of revenue, nearly half of revenue runs through Amazon, Google and Microsoft's cloud marketplaces, and there is roughly $637B of compute committed to partners. The founders also keep 50.1% of the vote through a single Class F share, so public investors are buying the economics, not the steering wheel. Ticker is ANT on Nasdaq, with a debut reportedly targeted for Thanksgiving week. Whatever you think of the valuation, this filing is the clearest evidence yet that the AI business model actually works.

This is so impressive and scary, how AI has become so smart. I love breaking things like this down, so here we go. OpenAI says its AI produced three new discoveries, in maths, computer science and physics, and handed them to some of the world's leading experts to check. The maths one is about waves. Think of dropping stones into a pond. Sometimes the ripples crash together and form one big spike for a split second. In 1991, a mathematician guessed that these huge spikes can never last long, and that most of the time waves are much calmer than at their wildest moment. It sounds obvious, but proving it has stumped mathematicians for 35 years. Humans managed to prove it for flat, two-dimensional surfaces in 2019. The AI claims to have proven it for every possible number of dimensions. As a bonus, this would also solve a famous puzzle called the Kakeya problem, which asks how little space you need to point a needle in every possible direction. Humans only solved that in three dimensions last year, and it was hailed as a once-in-a-century breakthrough. The AI claims to have solved it everywhere. The computer science result is the one getting the most attention. For 25 years, the field has assumed that certain puzzles are so hard that the time needed to solve them explodes exponentially, no matter how clever you are. This is the Exponential Time Hypothesis, and hundreds of papers in fields from biology to databases rely on it. The AI claims it is false. The two scientists who proposed it, Russell Impagliazzo and Ramamohan Paturi, are now leading the review alongside MIT's Ryan Williams. The physics one might be the biggest. Physicists have a rulebook for how the tiniest particles in the universe behave, called the Standard Model. It predicts experiments with incredible accuracy. But here's the strange part: nobody has ever proven that the rulebook actually makes mathematical sense. It's like a recipe that always produces a perfect cake, but no one can explain why it doesn't fall apart on paper. The AI claims to have proven that the rules for the force that glues the centre of every atom together are mathematically sound, along with a version of the whole rulebook. Part of this touches one of the seven Millennium Prize problems, a list of the hardest open problems in maths. A group led by Nobel Prize winner David Gross is now checking it. Why does this matter? The maths and computer science proofs, and the core of the physics one, were checked by a computer system called Lean that verifies proofs step by step. OpenAI isn't taking credit, and the experts reviewing it are unpaid, independent and free to say what they find. None of this is confirmed yet. But if even one of these holds up, AI has gone from helping scientists to making discoveries humans couldn't, in three fields at once.

Read the Anthropic filing tonight. Found something interesting. The global insurance industry collects about $8 trillion a year. Its costliest disaster ever, Hurricane Katrina, cost it $105 billion. Anthropic's IPO filing now warns that a catastrophic AI incident could be so big that the entire industry combined might not be able to cover it. Let that sink in.

Many of you don't understand how serious this is, so let me break it down. In computer science there is a famous kind of puzzle called 3-SAT. You are given a long list of yes-or-no switches and a set of rules about how they must be arranged, and you have to find a combination that satisfies every rule. The obvious way to solve it is to try every combination, but each new switch doubles the number of possibilities, so a puzzle with a few hundred switches would take longer than the age of the universe to brute force. For decades, the smartest people in the field found clever tricks that made this a little faster, but every trick still hit the same wall: the time required kept exploding as the puzzle grew. About 25 years ago, two computer scientists, Russell Impagliazzo and Ramamohan Paturi, working with Francis Zane, turned that wall into a formal idea. They proposed the Exponential Time Hypothesis, which says this explosion is not a failure of human cleverness but a law of nature. No algorithm, however brilliant, will ever solve 3-SAT without the time blowing up exponentially. Nobody could prove it, but almost everyone believed it, and it became one of the foundations of modern computer science. Researchers used it as a measuring stick. If you could show that solving some other problem quickly would also crack 3-SAT quickly, you had shown that problem was hopelessly hard too. Hundreds of papers, across scheduling, graphs, logic and much more, rest on this one assumption. Now an internal AI model at OpenAI claims to have broken it. It says it found an algorithm that solves these puzzles meaningfully faster than the hypothesis allows, and it has checked the proof in a computer verification system called Lean. And here is the remarkable part: the very people who proposed the hypothesis, Impagliazzo and Paturi, along with MIT's Ryan Williams, are now leading the public review of whether a machine has disproved their own idea. So what happens next? If the disproof holds, a large part of theoretical computer science has to be revisited. Every result that said "this problem cannot be solved faster, assuming ETH" loses its foundation, and researchers will have to work out which of those walls were real and which were simply where we stopped looking. It does not mean everything becomes easy. The new algorithm is still slow in absolute terms and this is not the famous P versus NP problem being solved. But it means our map of what is possible was wrong in a place we were very confident about, and it was an AI that redrew it. If the disproof fails, the hypothesis survives and the field breathes out, but we will still have watched an AI produce serious, checkable work on one of the deepest questions in the discipline, closely enough that the founders of the idea had to take it seriously. Either way, something has shifted.

In 2013, a 15-year-old high school sophomore emailed the SEC at 12:02 AM. He wanted to build his own stock exchange and register it as a National Securities Exchange. He had already spoken with an SEC representative, read up on the regulations and met a securities lawyer, and he asked to visit their Manhattan office. He even admitted it sounded naive. Twelve years later, the owner of the New York Stock Exchange invested $2 billion in his company. That kid is Shayne Coplan, founder of Polymarket. Be delusional.

update.

yeah, we are accelerating.

Anthropic just filed its S-1 to list on Nasdaq as $ANT. The numbers are wild. Revenue for the first half of 2026 was $16.3B, up about 13.5x from a year earlier. Q3 alone was reportedly $16.8B, so one quarter nearly matched the entire previous half. The loss looks apocalyptic at $80.7B in six months. But $72.4B of that is a non-cash accounting charge on convertible notes. The real story is compute. Anthropic has agreed to pay SpaceX about $3.1B a month for Nvidia-powered compute, up to $120.5B through 2029. The risk factors are unusual too. About a fifth of revenue comes from two customers, and the filing warns a catastrophic AI incident could exceed its insurance. The bet is simple: revenue has to keep compounding faster than the compute bill.

what did i just read????? anthropic has agreed to pay spaceX about $3.1B a month for Nvidia-powered compute, up to $120.5B through 2029.

posting the final message from the cancer patient we lost guys. this is the closure. life is beautiful. make the most of it.

i have never seen the consumer love for any product the way i see for aside browser by @AsideAI @hiddnest congrats to the team on building this incredible product.