Prakhar

Most "dark" bedrooms aren't dark enough. The threshold where measurable metabolic effects show up is lower than almost any standard sleep environment. Mason 2022 (PNAS, n=20): one night at 100 lux during sleep raised next-morning insulin resistance, increased nighttime heart rate, and decreased HRV vs <3 lux. Single-night, controlled. Obayashi 2020 (Sleep Med, n=678 elderly adults): bedroom light averaging ≥5 lux during sleep was associated with 3.74x diabetes incidence over 42 months. ≥3 lux still showed 2.74x. For context: 5 lux is roughly streetlight through curtains, or an LED display across the room. A hallway nightlight is 10 lux. Most people with any electronics, uncovered windows, or hallway bleed are above the threshold. Caveats: Mason is small and acute. Obayashi is observational, in elderly Japanese adults. Together: aligned, suggestive, mechanistically plausible. The lever (cover LEDs, close curtains) is mechanical and free. Mason, PNAS 2022: pubmed.ncbi.nlm.nih.gov/35286195/ Obayashi, Sleep Med 2020: pubmed.ncbi.nlm.nih.gov/31704511/

-Nicotine is a mitochondrial uncoupler -Increases testosterone and DHT Paired with aspirin and vitamin c cigarettes might actually be beneficial.

Scientists have created one of the most detailed 3D reconstructions of a human cell (eukaryotic cell) ever produced. This groundbreaking model, often termed a "Cellular Landscape Cross-Section Through a Eukaryotic Cell," combines data from X-ray tomography, nuclear magnetic resonance (NMR), and cryo-electron microscopy to map molecular structures in extreme detail.

Good sleep, good learning, good life super-memory.com/articles/sleep…

This is cool: maggieappleton.com/zero-alignment

Introducing Euphony, an open-source tool for visualizing chat data and Codex session logs. Paste in a public URL or upload a local file, and Euphony turns the raw data into an easy-to-browse view. It supports translation, filtering, editing, and more.

@aakashgupta @grok can you fact check these claims?

This is wild. SpaceX now has the right to BUY Cursor for $60B. Or pay them $10 billion to walk away. To put it in perspective, Cursor was worth $9.9 billion total in May of last year. Let's have a closer look at the numbers. Start with the $60 billion. Cursor was already raising money this week at a $52 billion valuation from a16z and Nvidia. The Elon offer sits 15% above a number that was already on the table. The next round priced in, with a one-year fuse. The $10 billion is the real number. That's what SpaceX pays even if it walks away and never buys the company. The walk-away fee alone is more than the entire company was worth 12 months ago. Now the strategic logic. Cursor stopped being just an editor in March. They shipped Composer 2, their own model, and it beat Claude Opus 4.6 on Terminal-Bench at one-tenth the price. The catch is that frontier coding models need frontier compute, and the only labs with frontier compute are the same ones building competing coding products. OpenAI shipped Codex. Anthropic shipped Claude Code. Google has Gemini CLI. Cursor was renting capacity from every company trying to kill it. Colossus is the way out. 230,000 GPUs in Memphis today, 1 million by year end, the biggest training cluster on Earth. The Information already reported Cursor is renting tens of thousands of those chips to train Composer 3. SpaceX is also building Grok Code, so they're not a clean partner. But xAI losing the coding race to Cursor is a better outcome for SpaceX than Cursor losing the coding race to OpenAI. The trade Cursor made: gave up the right to be acquired by anyone else for one year. Got training compute at a scale no other lab would sell them. Got $10 billion guaranteed if Elon walks. OpenAI tried to buy Cursor in early 2025 and got rejected. Cursor stays independent for at least 12 more months and gets to train on the biggest cluster on earth doing it. Elon just bought a one-year call option on Cursor for $10 billion. That's the deal.

Today, we’re open-sourcing the draft specification for DESIGN.md, so it can be used across any tool or platform. We’re also adding new capabilities. DESIGN.md lets you easily export and import your design rules from project to project. Instead of guessing intent, agents know exactly what a color is for and can even validate their choices against WCAG accessibility rules. Watch David East break down this shared visual language in action👇. New capabilities and links in 🧵

Audio Reactive LED Strips Are Diabolically Hard — Scott Lawson scottlawsonbc.com/post/audio-led

Today is my last day at Anthropic. I resigned. Here is the letter I shared with my colleagues, explaining my decision.

This 1 hour lecture on "Probability Theory" from MIT will teach you more about prediction markets than 2 month internship at at a Wall Street Quant firm. Bookmark this & give it 1 hour today, no matter what. It’s the most productive start you can give your week. Then read post below.

Claude users right now

A peanut-sized Chinese model just dethroned Gemini at reading documents. GLM-OCR is a 0.9B parameter vision-language model. It scores 94.62 on OmniDocBench V1.5, ranking #1 overall. For context, it outperforms models 100x its size. 100% open-source. It works in two stages. 1. A layout engine detects every region in a document. 2. Each region gets read in parallel. The model predicts multiple tokens per step instead of one. That's what makes it so fast at small size. It handles things most OCR tools struggle with: > Complex tables and nested layouts > Handwritten text and stamps > Math formulas and code blocks > Mixed image-and-text documents You can run it locally through Ollama. It fits on edge devices with limited compute. Every expensive OCR API just got a free competitor.

I've been playing with React Native WebGPU recently and it's really flipping my perspective on animations.

bro made a whip for claude 😭