Claude just computed a nine-loop scattering amplitude in planar N=4 super-Yang-Mills, past the eight-loop record. What that means. Scattering amplitudes predict how particles behave & physicists compute them in layers of correction called loops. Each loop makes the answer more precise & costs exponentially, sometimes factorially, it’s more work. Most real amplitudes have only been taken to two or three loops. The most precise prediction in particle physics, the electron's anomalous magnetic moment, used five. The challenge came from physicist Matt von Hippel, who publicly asked whether an AI could push past eight loops using only compute an academic could afford. Given one prompt & periodic instructions to continue, Claude ran largely unsupervised for days & did it, using bootstrap methods that Lance Dixon & collaborators developed. Total cost: a few thousand dollars. Dixon, who held the prior record, checked the result independently. Three caveats that matter more than the headline: 1. N=4 super-Yang-Mills is a toy model which is a testing ground for techniques, not a theory describing our universe. No new real-world particle prediction came out of this. 2. Claude applied existing human methods rather than inventing new ones. 3. A human-led group at the Chinese Academy of Sciences, using GPT-6 for parts of it, had concurrently obtained most of the same result & published a dataset on September 17. So the honest version isn't that a machine beats humans. It's that a hard calculation at the edge of a field now costs only a few thousand dollars & takes several days, by more than one route at once.

Been hooked on Octopath Traveler for days on my Switch. Gorgeous retro pixel graphics plus amazing background music. Each of the eight protagonists has a unique story, old-school turn-based fighting is super addictive.

At a family restaurant in Ohio, the waitress asked if I wanted soup or salad. I heard super salad. I did not know what made it super. But a man does not refuse the strongest option on the table. The super one, I said. Linda wrote it down without blinking and yelled toward the window. ONE SUPER SALAD! The cook leaned out. HOW SUPER, LINDA? HE DIDN'T SAY. GO ALL THE WAY. It came in a steel mixing bowl. Croutons like boulders. Two eggs. A whole breadstick planted in the middle like a flag. It should have cost eight dollars more than the soup. Linda charged me for the soup. There is a new line on the specials board now. SUPER SALAD. ASK LINDA.

New on the Science Blog: Yes, Claude can do Nine Loops. Theoretical physicists predict how particles behave using formulas called scattering amplitudes. These are notoriously hard to compute, so researchers work with layers of increasingly fine corrections called “loops”—each added loop makes the answer more precise but takes exponentially more computation. Most calculations stop at two or three loops. Eight loops was the previous record in a simplified model physicists use as a testing ground (planar N=4 super-Yang-Mills), set by SLAC's Lance Dixon and collaborators. Last month, physicist and science writer @4gravitons issued a challenge: could an AI push past eight loops in this model, using only the compute budget an academic could reasonably access? Given a single prompt describing the nine-loop problem, Claude ran largely unsupervised for days in Claude Science and solved it using methods developed by Dixon and his colleagues, at a total cost of a few thousand dollars. Dixon independently verified the result, and von Hippel wrote about the experience for our blog. Read more:

GPU-Backed Debt and the case for the End of SaaS Credit Shoal Signal Ep. 7 with @0xZergs, co-founder/CEO of @USDai_Official, hosted by @zaddycoin AI compute is being built faster than the financial system can finance it. Chips refresh every 24 months, and the debt that funds houses and planes needs decades of data and years to structure. David walks through why he thinks most of DeFi is money lending against money, and not real credit. 0:00 From art financing to GPU credit 2:18 DeFi is money market, not capital 3:13 Why CoreWeave bonds are not ABS 5:25 Convertible debt and the miner playbook 6:31 What securitization actually needs 9:03 Hyperscalers versus the Neo cloud tail 9:37 The depth perpetual idea 12:30 Underwriting eight chips, not companies 17:14 Why most RWAs are "toxic waste" 21:11 SaaS credit is dead 23:03 GPU loans as a super sector 26:37 Inside the Nvidia financing gap 30:20 Why only Nvidia chips 33:12 Compute as the next stablecoin 35:31 PYUSD, USDAI, and lowering the rate 38:34 Insurance with Barker and Munich Re 40:13 Internet capital markets versus margin 42:25 Where on-chain credit actually wins Youtube:

You need to understand that my father raised us with duty for our last name. The highest expectations because our last name is ‘Tate’. My Dad didn’t coddle me. He beat me when I fucked up, and I’m better for it. “I tell these boys the truth: life’s hard, stop crying, get tough”. That’s what a father does. This was beneficial to me. It prevented me from doing detrimental things in my life. He was a super principled man. He said I must analyse every firmly held belief I have, realize where it came from, who gave it to me, and whether they have my best interests at heart. He lived in a scary and combative world where he was surrounded by enemies in his mind. He was paranoid, but he was reacting to the reality he lived in. He had unfortunate incidents involving pure violence. He knew the world can become violent quickly. He told me that they are coming for me, like they came for him. I remember vividly, I was about eight or nine years old. My father won a chess tournament. A low level one. He made like $3,000, or something, and he said, “Listen, son, the rent is paid and I either want 100 grand or nothing”. So, we went to the casino and my Dad lost all the money. On the way back we went to a gas station. I went to the back to the coolers and I opened them to get a drink. Then I heard all this noise. When I turned around, my Dad was fighting three men. I don't know how the fight started. He had one of them up against the door and he was biting his face. The other two were punching my Dad in the back of the head. I'm a kid and I just stood there like, “What the fuck”, and the Korean guy behind is shouting, “Stop stop stop”. I didn't know what to do, so I went to walk over and although my Dad was getting punched, although he's biting this guy, I remember him turning to me and shouting, “Stay there”. Then he went back to biting the guy. I still don't know to this day how the fight started. My Dad had blood on his face. Some from the guy, some from him. His eye was black. His eye socket got fractured. His eyes closed and bruised up. We stood there and the Korean guy started panicking. The guys threatened to kill him as they left, “We're going to fucking kill you”, like gang bangers. Then he said to us: “My store is damaged. Wait for the police. Wait for the police”. I think my Dad was a bit disoriented. He never went down, but I could tell he was a bit dizzy. Dad stood there, in a military position, in the corner of the place, and waited for the police officer to come. The police officer came and started doing reports. He asked what happened. The police officer said to my Dad, “So you fought these three guys! Okay, well, what's your job”? My Dad said “I'm a chess player”, and the police officer said, “Well, maybe you should have been something else”. Then Dad said “My unmatched perspicacity coupled with my sheer indefatigability makes me a feared opponent in any realm of human endeavor.” I remember him giving this answer to the police officer. The police officer had a quizzical expression. His eyebrows were pulled down and together. And that was a saying that my Dad started using then on after, and I adopted it. That's the only time I ever saw him fight, that he didn't win. I mean, he got fucked up, but he didn't go down. He dealt with all three of them, and they ran, and he was left in the end. My father was a super unique individual. You have to imagine a big black guy. A physically dominating, dangerous guy who's a chess genius. He was also extremely condescending. But I understand why he was the way he was. When you're wired like he is, everyone else is a moron.

A video generation model can produce plausible clips — but can it reproduce the correct distribution of futures? These are very different things. Title: PAWBench: How Far Are We from Probabilistically Aligned World Modeling? URL: 🎯 Overview Just as a coin should land heads roughly 50% of the time over many flips, a video generator should reproduce physically correct outcome frequencies — not just look realistic. PAWBench is the first systematic benchmark for this "probabilistic alignment," evaluating 50 scenarios across 11 video generation models. 🔍 The Problem Prior evaluations focus on perceptual quality (FID, FVD) or diversity, but miss whether models generate each outcome at the right frequency (calibration) and whether they can produce all possible outcomes at all (coverage). PAWBench fills this gap. ⚙️ Evaluation Method PAWEval generates K=50 video rollouts, maps them to terminal outcomes via Gemini 3.5 Flash, and scores them using total variation distance (TVD) for calibration and valid-support recovery rate for coverage. Eight physical mechanism groups are tested: tossing, rotation, collision, material transition, and more. 📊 Key Results The best model (Cosmos 3 Super I2V) achieves TVD of only 20.5 (ideal = 0). The average TVD across all models is 31.2 — far above the 8.33 expected from finite-sample chance effects alone. Not a single model simultaneously achieves accurate probabilities, broad coverage, and high scoring reliability. ⚡ What Interventions Reveal Three approaches were tested: prompt engineering, coupled noise sampling, and LoRA fine-tuning. Each improved only partial metrics. Models consistently underreact to physically causal interventions and overreact to non-causal visual or textual cues. For research applying video generation to robotics, autonomous driving, or physical simulation, probabilistic alignment is now an essential evaluation axis. #VideoGeneration# #WorldModels#