Peter Thiel wrote a 23-year-old Microsoft employee a $74,000 check. The employee handed it back because he thought he might get promoted. That decision cost him $200 million. The employee was Garry Tan. Thiel had invited him to dinner because he wanted him to join a startup he was funding. Tan had just graduated from Stanford and was working as a low-level product manager at Microsoft. Thiel asked how much he made. “$74,000.” So Thiel wrote him a check for exactly that amount. “Cash this check, quit your job, and let’s go change the world.” Tan says he was sitting there over appetizers when he gave his answer. “Thank you very much, Mr. Thiel, but I might get promoted to Level 60 next year.” The startup was Palantir. Thiel had offered him about 1% of the company. Tan eventually joined as employee number 10. But he calls turning down that first offer his $200 million mistake. How many people have chosen a promotion over the chance to own part of the future?

Frankie Muniz had 38 cars before he turned 18. Then his girlfriend found a $41,000 charge for “research” that he never made. “It says you spent $41,000 on research last month.” “I didn’t spend $41,000 on anything.” “Call him right now and ask.” So Muniz called his business manager. He says the man immediately got “really frazzled.” Muniz had grown up without money. Once Malcolm in the Middle took off, one of the first people hired to manage his new life was this man. “I really, really trusted him,” Muniz says. But that $41,000 call made him start going through years of records. He says he found expenses that made no sense, car insurance far higher than the cars were worth, and deals involving the manager’s friends. The person trusted to protect a child star’s money had been handling it while the child was too busy becoming famous to question every line. Muniz says the documents eventually filled banker boxes. But by the time he understood what had happened, the statute of limitations had expired. Police couldn’t help. He had 38 cars before he was 18. And nobody had asked why he was supposedly spending $41,000 a month on “research.” How many child stars looked rich enough that nobody noticed they were being robbed?

QWEN IS NOW RUNNING IN A WEB BROWSER AT UP TO 500 TOKENS/SEC Developers just released compressed versions of Qwen3.5 that run entirely locally through WebGPU The 2B version is just 430MB and reportedly hits up to 500 tokens/sec on an M4 Max They also released 4B and 9B versions, with the 9B model's language weights taking up just 2GB At this rate, you won't need a $10,000 PC just to run open-weight AI models locally Demo's link in comms

I gave Muse Wonder access to my PC and Blender and told it to make an Optimus Prime model I call this AI cuckolding I’m sure it could’ve done a better job, but I needed my PC to watch the new season of The Apothecary Diaries It spent three hours making this, stopping for several minutes after every single action🙄 Here’s the result

the iphone has a touchscreen the AI has no fingers so someone gave it a $7 ESP32-S3 it watches the screen through AirPlay, sends taps and keystrokes via USB HID, and lets a vision model decide what to do next no jailbreak. no computer needed once it's set up

Asked Gemini 4 Argon to visualize Earth’s entire 4.54-billion-year history. My main requirement: scientific accuracy. Then I asked it to break down what the visualization is actually based on: >Present day - GSHHG coastlines and NOAA ETOPO 2022 elevation on a 60-arc-second grid (~1.85 km between grid points at the equator). NASA Blue Marble surface imagery is from 2004. >The last 80,000 years - ice-sheet data from the PaleoMIST 1.0 reconstruction, at 2,500-year intervals. The steps between those snapshots are interpolated. >540 million years ago to today - PALEOMAP / Scotese & Wright paleogeography, interpolated between map snapshots. >1.8 billion to 540 million years ago - CAO2024 continental-block reconstruction. Ancient positions are uncertain; coastlines and relief are illustrative. >Earlier Earth and the Moon-forming impact - visual reconstructions, not a literal geophysical simulation. The further back in time you go, the more the visualization relies on reconstruction rather than direct geographic data. misatodevx.github.io/earth-d…

Mark Cuban says his receptionist stole $82,000 of the $84,000 his company had in the bank. When the bank called, he thought they’d caught her in time. “You didn’t cash them, did you?” “Of course we did. We’re a bank, sir.” This was at MicroSolutions, his first company. The receptionist was supposed to put checks for their suppliers in envelopes and take them to the post office. She used white-out to cover the recipients’ names, wrote in her own, and cashed the checks at the bank’s drive-through. The company had $2,000 left. The suppliers were still waiting for their money. Cuban and his partner had to call them and explain what had happened. They asked for time and promised they would get paid. The suppliers agreed to work with them. The business survived. In 1990, Cuban sold MicroSolutions to CompuServe for $6 million. Shannon Sharpe asks whether they pressed charges. Cuban says they couldn’t find her. As far as he knows, nobody ever did. He’s a billionaire now, but when he tells this story, he still spells out her name. I can understand why. The bank called him about checks with the recipients’ names covered in white-out, then explained that it had already paid them. Would you have trusted that bank again?

the new meta model is getting so much hate… keep it coming, guys😉 muse wonder isn’t a bad model in itself, but it’s clearly no Astra - let alone Opus for this test, I built a procedurally generated 3D game using Three.js. As you can see, it looks pretty good misatodevx.github.io/afterho… it took half a day of back-and-forth prompting to get this result. Opus 5.5 would’ve done it in three prompts - and probably better

Anthropic has research engineering roles paying up to $850,000 a year People in these positions don’t just know how to write prompts for models. They understand how neural networks work One of the best ways to learn about this has been free on YouTube for years Of course, I don’t think this will get you a job at Anthropic or OpenAI, but it’ll put you 10 steps ahead of 99% of other AI users Nando de Freitas taught this course at Oxford before joining DeepMind. It predates the current AI boom, which is exactly why it covers the fundamentals I suggest spending a couple of evenings without Netflix. I’m sure you won’t regret what you learn

A garden on Mount Olympus… in Three.js?? Koi swimming beneath marble arches, butterflies fluttering around the trees, waterfalls pouring straight into the clouds. Remember when we used to search the internet for something mesmerizing to look at? Now we can get exactly what we want with a single prompt. Made with Opus 5.5, by the way.