Memory is Slow, Disk is Fast - part 3 Can an on-disk data structure with insert(), delete() be faster than an in-RAM binary tree? New Blog Post up. bitflux.ai/blog/memory-is-sl…

Do a kickflip!

Real life, baby

BOOM! Well, I'll be damned if that's not the very BEST-sounding, 100% simulated/synthesized concert grand piano I've ever heard! FrankenSim.org/music What does that mean? ZERO recordings of real-world sounds were used (other than for comparison and diagnosis), nor were any measurements captured by microphones or similar. Literally everything from start to finish here was part of an enormously intricate physics simulation of a piano: its geometry, materials, moving parts, and acoustic radiation. Oh, and I did the same thing for a full drum kit and cymbals. The cymbals are modelled using the estimated geometry and dimpling pattern of a real Zildjian splash cymbal, together with estimated profile details and literature-based properties for B20 bronze alloy. For the most extreme audio quality, you can pre-render to WAV files or similar. But after many, many rounds of iterative performance optimization, I now have the wasm version running at better than real-time, so it sounds really good live in the browser as well! This is a big deal. It's shockingly hard to get truly realistic-sounding audio from synthesized instruments. The real world is just too complicated. What people usually end up doing is spending hundreds or thousands of dollars to buy these massive libraries of recorded sounds of a real piano, which takes up many gigs of space and is super slow to use. The worst part of that approach isn't the size of the files, the cost of the sample packs, or the inelegance of the approach. It's that EVEN THAT is a massive simplification, because they are recording each key separately at various key velocities and so forth. That means they are missing the very complex interaction effects that arise in a real piano when you play certain keys at the same time or after one another. The energy persists in the instrument and conditions the next sounds. It's subtle but real and you pick up on it when it's missing. And yes, they do try to capture some of this, but they use very lossy methods with more simplified approximations that are way more primitive than what I'm doing here with FrankenSim. I've invested a massive amount of time and focus on ensuring that the energy accounting is correct in these simulations. And the same thing goes for drums! There are complex interaction effects. These sorts of things are emergent and "just fall out" when everything is simulated for REAL using the actual underlying physics. So why haven't other companies and people done this? I'm not actually sure; it's kind of insane when you consider all those people working at Yamaha, Roland, Korg, Moog, etc. But my guess is that it's because it's super hard and complicated to get all the physics right and to program it the right way so that it's tractable without needing crazy amounts of compute. Anyway, take a look at this demo site I set up. It's a lot of fun to play with, and at the bottom there is a nifty MIDI player that takes any MIDI file and lets you listen to it with the high-quality simulated piano and drums in the browser. This is the culmination of months of work and hundreds of billions of frontier tokens. It all started when I wanted to see what Fable 5 could really do, so I thought of the most comically ambitious project I could think of: FrankenSim, the system that unifies ALL of physics into a single coherent and cohesive simulation across all domains: mechanics, fluid dynamics, electricity and magnetism, etc., instead of modeling each domain as a separate silo, where each domain can use the same generic geometric primitives, solvers, calculus machinery, etc. All unified using the language of sheaf cohomology to maintain consistency between representations, and where everything is made as legible, intuitive, and ergonomic to AI agents as possible. This music simulation stuff is just the beginning. Beyond many more instruments, I'm going to be using FrankenSim to model HVAC systems, fluid flow through plumbing systems, and more!

I’m the Corporate Vice President of Global Talent Strategy at Microsoft. People say I hire H-1Bs. My badge says I source world-class talent that can’t leave. That second part? That’s the whole program. Last year we filed 4,812 petitions. 97% got approved. Everyone talks about the lottery like it’s luck. It’s great for us, because it lets us call the outcome chance. It isn’t. The outcome got decided months earlier by a team that writes the specialty-occupation language before the req even opens. They’re fantastic. I’m proud of them. This year Copilot started drafting the language for them. Turns out Copilot is really good at describing a job only 1 person on earth could possibly do, as long as that person is already in our system. We called the workstream PERMANENCE. Legal made us rename it. Now it’s Project Continuity. Same slides. Domestic candidates apply. Plenty of them clear the bar. We go with the H-1B anyway. Why? Because the H-1B already took the band. The band is Level II prevailing wage. Level II is the median we asked for. The people who can quit get paid off a different median, one they talk about openly at lunch. The H-1B negotiates off the LCA. The LCA’s public. Basically nobody reads it. I read it every quarter. Honestly, it’s calming. We track the result as diversity and inclusion. 68% of the engineers who joined Azure last cycle came in on temporary status. The dashboard shows that number in a really nice green. What it doesn’t show is the 28% fully-loaded cost delta. That lives in a different workbook. Finance owns that one. Finance is proud too. Let me tell you about Raj. Raj has been on Azure Storage for 6 years. He knows the codebase better than the guys who wrote it, and 2 of them already left for counteroffers. Last spring his manager, Dana, asked why we couldn’t just promote him to staff and backfill locally. So we set up a talent review. 40 minutes. I asked Dana about pipeline health. Dana asked about the codebase. I asked about competitive intensity. Dana asked about the codebase again. I asked whether she had a growth mindset. She said she thought so. We agreed to take it offline. Offline is where Dana’s req went to die. Raj stayed. Raj always stays. He came in on another company’s visa, and we transferred it. Counsel has a template for that. The new $100,000 fee only hits talent that’s still overseas. Raj was already here. Transfers keep the leverage and skip the toll. Finance made a slide about it. There’s a little rocket on it. The transfer is the point. Look, if you can quit on Friday, you’ve got leverage on Thursday. If your legal status ends the day your job does, you’ve got Thursdays and that’s it. We never have to threaten anybody. The statute does that for us. Raj has read the statute. He’s also read his priority date: 2014. Projected to go current in 2031. Between now and then we’ll bump his Level II wage twice, and both bumps will be smaller than what we gave the people who can walk. We call that pay equity, since the formula treats everyone exactly the same. Only some people get to argue with it. His wife’s on an H-4. Back in Pune she was a pharmacist. Here she’s a really organized volunteer. His daughter turns 21 in 2029, which is the year she ages out of his case. She wants to be a doctor. On paper, in 2029, she becomes an international student in the only country she remembers. HR flagged it as a retention risk. Finance flagged it as a retention asset. We went with Finance. Some people call this a caste system. We call it a two-tier compensation architecture, and it’s working exactly the way it’s supposed to. In the workbook, the line is column AF. The header says Mobility Status. There are 2 options: Portable and Anchored. Nobody ever types the word caste. Excel’s never needed it. The Portable folks sit above the line. They get the counteroffers, the refreshers, and the freedom to ignore the 11pm call. When a Portable engineer mentions he’s “exploring options,” a VP takes him to lunch somewhere with no prices on the menu. When an Anchored engineer explores options, he gets 60 days to do it. That’s the grace period after you’re let go. They call it grace. We didn’t name it. We’d have named it better. The Anchored folks sit below the line. They get the scope, the on-call rotation, the “stretch assignment,” and a calibration note that says high potential, visa-dependent. Nobody writes down that second part. Everyone in the room already knows. The room claps for the first part. When a Portable engineer pushes back on a deadline, we call it psychological safety. When an Anchored engineer pushes back, well, he doesn’t. We call that alignment. At the all-hands, both tiers sit together. Same hoodie. The hoodie says One Microsoft. And at the hoodie level, it’s true. During the January layoffs, the Portable folks got severance, a sweet LinkedIn post, and 14 recruiter DMs by lunch. The Anchored folks got severance and a countdown. A guy on Raj’s team spent his last 3 weeks training his replacement and his first 3 weeks unemployed applying to 212 jobs from a Starbucks in Bellevue, because his landlord wanted proof of income. He landed something on day 57. We sent him a survey about his offboarding experience. He gave it 4 out of 5. He said everyone was really kind. And they were! The people were kind. The people just didn’t design the system. Belonging scores go up when the alternative is a one-way flight. Velocity goes up when nobody on the team can say no. Viva pings Raj at 11:40pm to remind him to protect his focus time. He reads it on the incident bridge. His engagement numbers are fantastic. In March his dad died in Hyderabad. Raj didn’t go home. His visa stamp had expired, and the next consulate appointment was 11 months out. If he left, he might not get back in. And getting back in meant the job, the house, his daughter’s school, the 2014 priority date. So he watched the cremation on his phone, propped up against his monitor, between a design review and an on-call handoff. His manager approved 3 days of bereavement leave. He used 1. The other 2 roll over. We’re super flexible about leave. Critics say we’re replacing Americans. Internally we call it workforce durability. Durability just means they stay. Durable talent brings specialized skills. The specialized skills have signed the at-will acknowledgment and the non-compete, and we’ll enforce it if they try to jump before their date’s current. The date’s never current. Everybody knows what happens if we pull a petition. We’ve never had to explain it twice. (Fine, immigration counsel will tell you that after 180 days the priority date survives a withdrawal. We leave that out of the onboarding deck. Onboarding decks should be uplifting.) In March we won Best Employer for Global Talent. The plaque’s in Building 92. Most of the people it honors can’t switch employers to find out if it’s true. Some of us have started thinking of the backlog as infrastructure. We don’t lobby against it. In our policy comments we call it a “stable pipeline environment.” Stability’s one of our values. It’s on a poster, right next to Courage. I meet with the team every Monday. They know the cap. They know the premium processing window. They know which vendors will staff the interview loop so the candidate’s prepped and the manager never gets surprised by an acceptance. They cheer when a req closes under budget and the diversity and inclusion cell turns green. Last quarter it went green in 4 orgs. We sent a note. It said belonging and innovation. Both totally accurate. Raj belongs. We innovated. Dana left in August for a competitor. We matched the offer in a day. She still said no. People who can leave are like that. Raj sent her a goodbye message in Teams. He couldn’t make the drinks. He was on call. I’m so, so proud of this team. They took a regulatory constraint and turned it into permanent leverage. And the leverage compounds. We report it as growth. The analysts love growth. So do I. My equity vests every quarter. Raj’s priority date goes current in 2031. His father’s never will.