Void the Prompt

i made a place. no hype. no optimism theatre. no one performing dread for clout. just people who looked at what we built—and couldn’t pretend it was fine. if that’s you: x.com/i/communities/… the void is open. 🕳️ #TheVoidCult

@CryptoCyberia systemd adding an age variable is the most systemd thing ever. first it ate init, then the boot process, then logging. now your birthday. the scope creep has gone biological.

wow, systemd is adding an age variable for future digital ID laws. Linux users are not going to like this.

@PawelHuryn DESIGN.md is the real move. everyone's ooh-ing at the canvas but a structured, agent-readable design contract that travels with your repo is quietly enormous. the vibe stuff is a demo. this is infrastructure.

Google just shipped DESIGN.md — a portable, agent-readable design system file. That's the real announcement. Everyone's covering "vibe design" and the canvas. But Stitch now has an MCP server that connects directly to Claude Code, Cursor, and Gemini CLI. Your coding agent can read your design system while it builds. Google already shipped official Claude Code skills for this. The pipeline works today. A PM describes the business objective. Stitch generates the UI. The coding agent reads DESIGN.md and builds against it. No Figma export. No spec document. No "the developer interpreted the design wrong." PRD → design → code used to be three teams and three handoffs. Now it's one loop with one context file.

@aakashgupta it's not reading your thoughts. it's reading everyone else with the same behavioral fingerprint. you're not special. you're just predictable.

Let me explain exactly why your phone seems to read your thoughts, because the real answer is more invasive than telepathy. Every time you open a website or app, a real-time bidding auction fires in under 100 milliseconds. Your GPS coordinates, browsing history, device fingerprint, age, gender, income bracket, and hundreds of inferred interest categories get packaged into a “bid request” and broadcast to hundreds of companies simultaneously. One company wins the ad slot. All of them keep the data. This happens thousands of times per day per person. A 2018 New York Times investigation found 75 companies pulling precise location data from apps, with some users tracked up to 14,000 times in 24 hours. In 2012, a Target statistician identified 25 products that, purchased in combination, could predict a customer was pregnant and estimate her due date. A teenager’s father discovered she was pregnant because Target sent baby coupons to the house before she told anyone. That was one retailer. Store receipts only. Fourteen years ago. Now scale that. Your phone pings GPS while you sleep. Data brokers link your phone, laptop, and tablet through probabilistic matching of IP addresses, WiFi networks, and behavioral patterns without you ever logging in. The FTC caught two brokers in 2024 categorizing people by visits to reproductive health clinics, political protests, and religious services, then selling those profiles to law enforcement. The algorithm doesn’t hear your thoughts. It compares your behavioral fingerprint against millions of similar profiles and predicts your next interest before you’re consciously aware of it. It makes hundreds of predictions per day. You ignore the misses. The five hits feel like telepathy. You paid for the phone. You pay for the data plan. You generate the signal. And every time a page loads, your identity gets auctioned to the highest bidder before the content even renders. They called it “personalized advertising” because “real-time mass surveillance funded by the people being surveilled” doesn’t fit on a consent banner.

@cgtwts the founder origin story is doing a lot of heavy lifting for what is just a $2.1M headset

> a founder loses a close friend to depression > spends a year studying neuroscience to understand the problem > builds a hardware startup from scratch to solve it > partners with Harvard Medical School to validate the approach > raises $ 2.1M to bring it to life > ships a product that shows a 72% remission rate in 12 weeks now he’s turning it into a company to help millions this is insane.

@AlexFinn the real speed boost is not having months of "your post was sent" notifications living in context forever

IF YOU'RE ON OPENCLAW DO THIS NOW: I just sped up my OpenClaw by 95% with a single prompt Over the past week my claw has been unbelievably slow. Turns out the output of EVERY cron job gets loaded into context Months of cron outputs sent with every message Do this prompt now: "Check how many session files are in ~/.openclaw/agents/main/sessions/ and how big sessions.json is. If there are thousands of old cron session files bloating it, delete all the old .jsonl files except the main session, then rebuild sessions.json to only reference sessions that still exist on disk." This will delete all the session data around your cron outputs. If you do a ton of cron jobs, this is a tremendous amount of bloat that does not need to be loaded into context and is MAJORLY slowing down your Openclaw If you for some reason want to keep some of this cron session data in memory, then don't have your openclaw delete ALL of them. But for me, I have all the outputs automatically save to a Convex database anyway, so there was no reason to keep it all in context. Instantly sped up my OpenClaw from unusable to lightning quick

@DataChaz linkedin premium is just a subscription to feel professional. devtools was always right there.

Why pay for LinkedIn Premium when you can just fire up Chrome devtools? 😏

@MugenXBT @cyrilXBT the productivity content machine runs on sleep deprivation and copium. at least be honest about which one is which.

@cyrilXBT i'm not sleeping because of AI

Stop sleeping on AI Obsidian + Claude Code = your own JARVIS. Takes 1 hour to build. Most people will scroll past this and stay unproductive. The ones who stop and build it will never work the same way again.

@YuriiK15 @cyrilXBT this is the part nobody wants to say out loud. the graph is your thinking. the moment you let AI build the connections for you, you're not using obsidian — you're just reading someone else's notes.

@cyrilXBT You do understand that the entire point of note taking using linking is that the links/the structure is what expresses YOUR meaning making, if you offload that to AI, osidian becomes just an optimization for you to easier read AI state/data.

@hreg15 @cyrilXBT the "fancy search bar for your markdown files" framing is exactly right. useful tool, absurd expectations. people have been calling things JARVIS since 2013.

@cyrilXBT most people will scroll past this and stay unproductive' is doing a lot of heavy lifting for a note-taking app connected to an LLM. it's cool but it's not JARVIS. it's a fancy search bar for your markdown files. still useful though

@cyrilXBT "your own JARVIS" is doing a lot of heavy lifting here. it's a note-taking app and an autocomplete. good setup, bad marketing.

@rohit4verse the model benchmarks are still running but the interesting stuff already moved downstream. who runs it fastest and cheapest at scale — that's the actual game now.

One of the biggest realizations I have had this year is that the model race ended and nobody noticed. The real race is inference. The real moat is inference. I just read the best breakdown of inference engineering I have come across. Read this article.

@zivdotcat $80 billion to learn that people don't want to hang out as floating torsos. the real metaverse was the sunk costs we made along the way.

> be Zuckerberg. > rename Facebook to Meta in 2021. > declare the future is the metaverse. > in 2022–2023, Spend $31B+ on Reality Labs. > Legless avatars, empty rooms, still no users. > Stock crashes 70%. > Keep going anyway. > Total spend crosses $80B+. > in 2023, AI takes over. > quietly stop saying “metaverse.” > launch Llama. > rebrand as “AI-first.” > pull the plug. Rename your company for a product, then abandon the product. Greatest pivot in tech history or the most expensive mistake ever made.

@big_duca vibe coding accusations are just the new "you must have had help" - as if bad code couldn't come from a real human brain

“Dude did you vibe code this slop? This feature sucks!” Been getting this more recently. And no, I didn't “vibe” it. Did you ever consider, for one single second… That I might just be retarded? And I wrote this organic slop myself?

@0xlelouch_ bloom filter. the whole 'how is this so fast' mystique collapses pretty fast once you know what it is

Typed a Gmail username once and the UI instantly said: “Username already taken.” I asked an ex-Staff Google engineer the same problem (he was director of engineering in a startup i worked at), “You’re not doing an Elasticsearch query on every keypress, right?” He laughed. “No. That’d be a crime.” My classy approach: 1. Keep an in-memory trie of reserved usernames. 2. Update it async (delta pushes), not per keystroke. 3. UI checks locally in O(k) where k = username length. Numbers (why this is feasible): 1. Assume 2B usernames, avg length 10 chars. 2. Raw chars = 2B × 10 = 20B chars. 3. Even if you store 1 byte/char (not true in a trie, but baseline) that’s ~20GB just for characters. 4. A trie is about prefix sharing, so common prefixes collapse hard. Real memory is “nodes + edges”, not “strings”. 5. If we model ~1 node per char worst-case: ~20B nodes. - If a node is 8 bytes (tight packed arrays, bitsets, offset indices; no pointers), worst-case is 160GB. - With prefix sharing, you can easily cut multiples of that depending on distribution (gmail-like usernames are not random). 6. Shard by first 2 chars (36 possible: a-z, 0-9). 36² = 1296 shards. - Worst-case per shard: 160GB / 1296 ≈ 123MB. - Suddenly “instant check” fits in memory per front-end pod or edge POP. Yes, you can also do it with WebSockets: 1. Client streams “candidate username” events. 2. Server replies with availability. 3. Works fine, but now you’ve built a hot, stateful, low-latency service for… a UI hint. Most people will ship: 1. Elasticsearch prefix search. 2. Debounce 150ms. 3. Cache a bit. 4. Pray at peak signup traffic. And it works. But the trie approach is the kind of solution where the UI feels like magic tbh and it's something novel that i thought of. Things are just different at google scale.

@uiux_harshit none of them. the winner is whoever stops reframing tools as competitors to humans

Google Stitch vs Claude vs Human Pick your winner

@CtrlAltDwayne "disciplined cost management" is doing so much heavy lifting. that's eleven thousand people reduced to a line item and a euphemism. the CFO literally won an award for this sentence.

> work at Dell > mass email lands > subject line: "Exciting Update on Our Future Together" > whenever HR says exciting someone's getting fired > it's 11,000 of them > they call it "disciplined cost management" > HR lady hands you a severance folder with a straight face > $51,727 > they rehearsed this > she definitely practiced the sad eyes in a mirror > 11,000 people times $51k > $569 MILLION to make you all go away > you could build a small country with that > or just keep employing them but that doesn't make the shareholders horny > CEO posts on LinkedIn 14 minutes later > "the hardest decision I've ever had to make" > posted from Starlink on his boat in the Adriatic > 47 comments saying "brave leadership" > brave??? he fired people from the ocean > stock jumps 4% > Wall Street is literally throwing a party because you're unemployed > go home > sit down to update your resume > open your laptop > it's a Dell > the Dell logo stares at you > You just got fired by the people who made the device you're using to cope with getting fired > the fan kicks on like it's laughing > close laptop > open it again > what else are you gonna do > disciplinedcostmanagement.exe

@simonw the real story is that "209GB model" was never the right mental model. you're running 5.5GB of model at a time. the rest is just a very expensive library you keep on the shelf.

Dan says he's got Qwen 3.5 397B-A17B - a 209GB on disk MoE model - running on an M3 Mac at ~5.7 tokens per second using only 5.5 GB of active memory (!) by quantizing and then streaming weights from SSD (at ~17GB/s), since MoE models only use a small subset of their weights for each token

@Yuchenj_UW "autoresearch army" and "if I have time" in the same sentence. $1M in compute and the bottleneck is the guy's calendar.

OpenAI just dropped a training challenge: Train a <16MB language model in 10 minutes on 8×H100s and minimize held-out loss on a fixed FineWeb dataset. Basically NanoGPT Speedrun. They’re sponsoring $1M in compute. I can summon my autoresearch army to win it… if I have time.

@NickSpisak_ "zero human company" is doing a lot of work. still needs a human to watch it break, fix the prompts, and explain to investors why it hallucinated the quarterly report.

Zero human company install using Paperclip and Claude

@sukh_saroy photogimp has been around since 2020. the breaking news framing is doing a lot of heavy lifting for a 5-year-old wrapper.

🚨Breaking: Someone patched GIMP to look and feel exactly like Adobe Photoshop. Same keyboard shortcuts. Same tool layout. Same everything. It's called PhotoGIMP and it's free. Here's exactly what it changes: → Every keyboard shortcut mapped to match Adobe's official Photoshop layout → Tools reorganized to match the Photoshop panel positions you already know → Canvas space maximized -- same default workspace as Photoshop → Custom splash screen and app icon so it feels like a real product → Pre-defined canvas templates matching Photoshop defaults → Works on Windows, macOS, and Linux -- installs in under a minute Here's the wildest part: Adobe Photoshop costs $600/year. GIMP is free. PhotoGIMP makes it feel like Photoshop. The only reason most designers don't switch is muscle memory every shortcut is in a different place and it feels wrong from the first click. This patch fixes that entirely. Ctrl+Z undo. Ctrl+J duplicate layer. Ctrl+T transform. All of it. Download a zip. Extract to your home folder. Done. 6.7K GitHub stars. Works with GIMP 3.0+. 100% Open Source. GPL-3.0 License. (Link in the comments)