Chinmay

EUV machines are the most complicated tools humans make. Their supply chain has over 10,000 individual suppliers, and any one of them not scaling fast enough can bottleneck the entire AI industry. An EUV tool fires lasers at a tiny tin droplet three times in precise sequence, blasting it hard enough to emit EUV light. That light bounces off 18 multilayer mirrors onto the wafer. Meanwhile, the two platforms inside the machine - one holding the stencil, one holding the chip - are flying back and forth at 9Gs in opposite directions. The successive passes have to land on top of each other to within 3 nanometers. If any part of this is off, yield goes to zero. Take just one component. The mirrors are mostly supplied by Carl Zeiss, who have probably fewer than a thousand people working on them. In turn, Carl Zeiss rely on machines from Switzerland to deposit each of the layers, and use a coating process co-developed with a different German company. None of these companies have woken up. They’re gradually increasing production, but nowhere near the levels necessary for what the labs want by the end of the decade. @dylan522p predicts production can't scale beyond about 100 EUV machines per year by 2030, no matter how much money gets thrown at the problem. In the medium term this is the key bottleneck on scaling.

reveal.js 6.0 is out! ⚡ New Vite build system 📦 Now ships with TypeScript types ⚛️ Official React wrapper (@revealjs/react) 🌚 Dark mode for revealjs.com Full changelog and upgrade guide: github.com/hakimel/reveal…

log (😅) = 💧log 😄

There are currently two mature @TC39 proposals for the low-level functionality that would make it possible to declare *any* file type as a dependency: 1. Import text (Stage 3 as of last week!): github.com/tc39/proposal-… 2. Import bytes (Stage 2.7): github.com/tc39/proposal-… Perhaps we should push for these to get implemented *first* instead of ad hoc types for every possible type of dependency (CSS modules, HTML modules, etc), which doesn’t scale. Especially Import bytes, which is the root of them all: Even Import text can be trivially implemented on top of it via `new TextDecoder("utf-8").decode(bytes);`

One underrated skill that more engineers should learn is the "ability to follow up" and "escalate when things are beyond you". A lot of times I've seen engineers will raise a request for an access or ask for a PR review. Days would pass, and they would not even follow up once. They assume that - since I have requested for access, or I have requested for a review, my job is done. Your job is to get work "done", not play ping-pong. So in case you are blocked on something or someone, learn to follow up and also escalate if things are not moving forward beyond a certain time. I get that in the ideal world, the other person will approve your request or review your PR in reasonable time. But if it's not happening, the problem is still yours. You are still blocked, and if you are blocked, the ownership to get unblocked is still yours. A lot of high agency folks operate that way. Learn the art of following up and escalating things when you have done your job. You'll go far in your career this way.

for a while i've had a slight fear that the bluetooth from my airpods could be frying my brain this weekend i pulled the raw data from a $30m government study of 1,679 mice blasted with cell phone radiation and reanalyzed it what i found was...not what I expected? 🧵

If you are using MCP to access your data, you are trusting that third party with unencrypted access to your data. Third-party MCPs are not an option if you value privacy. End-to-end encryption makes MCP effectively useless.

Ooooo that's a shitty dilemma about TrueNAS

The world is a museum of passion projects

Somewhere out there is a guy who uses Notion, Superhuman, OpenClaw on a Mac Mini, Raycast, a mechanical keyboard ($400), Wispr Flow, and gets nothing done every day

(cont'd) This week, for our SLAM course, I built an interactive lab to visualize SE(2) odometry uncertainty propagation — covariance ellipses, noise tuning, and ground truth, all in real time. 🤖 dgist-slam.github.io/se2-diff-odom-…

T3 Code is now available for everyone to use. Fully open source. Built on top of the Codex CLI, so you can bring your existing Codex subscription.

Defuddle now has a website! This means you can use Defuddle anywhere to get the main content of a page in Markdown format. You can simply add "defuddle.md" before any URL, use it via curl, Skills, CLI, or add it to your app via NPM. x.com/kepano/status/…

This "book filter" idea sounds interesting to me. "Is anything in this book new to me?" I have many years of notes and highlights that I could load into NotebookLM (or, for that matter, a project in ChatGPT or Claude). Then, upload a book and ask it to just show me parts that are not yet captured by my notes.

@rhatr I don't do it because it's a pita to type it in terminal, for no apparent reason. Idk why just double quotes and spaces doesn't work lol.

I've just seen the worst enum in my life

we desperately need a new season of silicon valley. the ai era alone would carry 3 seasons

This story is actually insane: • dude drops $2000 on a DJI robot vacuum like a lunatic • refuses to use the normal app like a peasant • Sammy Azdoufal fires up Claude to crack the API so he can drive it with an xbox controller • Claude delivers the goods • pulls an auth token from their servers, connects successfully • except the system thinks he controls 7000 vacuums • checks again • yep, seven thousand • DJI built authentication with zero device ownership verification • any valid token works for any unit on the planet • Sammy now has eyes inside homes across 24 countries • live vacuum camera feeds everywhere • full floor plans from the mapping data • some guy in germany eating cereal at 3am, unaware his roomba is snitching • one API call away from being the most informed burglar in history • all he wanted was to steer his vacuum with a joystick • does the right thing and reports it • DJI fixes it in two days • back to normal life with his stupidly expensive floor cleaner • IoT companies stay undefeated at shipping garbage security

Lots of talk about how the detachment from code inherent in AI code generation tools reduces code quality. But what about the ways it *improves* code quality? This detachment can be beneficial too: it reduces emotional attachment to the code artifact, and the biases that come with it. For example, a well-documented cognitive bias in creative work is the “IKEA effect”¹ where creators overvalue artifacts they created. The more time we spend creating something, the more attached we become, and the harder it is to see its flaws clearly. Painters are often advised to periodically turn a canvas upside down, precisely to break that attachment and regain objectivity. AI reduces exposure to the code artifact being produced, which obliterates this bias: 1. It inserts a layer of indirection between the creator and the artifact. You are no longer engaging with the code primarily as its author, but as a reviewer and mentor. You may edit, refine, and restructure it, but your default stance shifts from “I made this” to “Does this hold up?”. 2. Bias increases with time and effort invested. AI dramatically reduces both the time and effort sunk into any given implementation. Lower sunk cost makes it easier to delete, rewrite, or discard. Another positive effect is reduced rejection sensitivity. When teammates review “your” code, it can feel personal. But for a PR "co-authored with Claude", they are effectively joining you as co-reviewers. Change requests are no longer critiques of your craftsmanship; they’re corrections to an external artifact you are all evaluating together. The concern that detachment harms quality assumes an idealized human, perfectly objective and rational — a standard no actual human could ever meet. ¹ en.wikipedia.org/wiki/IKEA_effe…