Gul Saeed Khattak

𝗜 𝗕𝘂𝗶𝗹𝘁 𝗡𝗲𝗿𝗱𝗟𝗼𝗿𝗲: Karpathy Built HackerTube. I Took It One Step Further. Last weekend Andrej Karpathy opened YouTube late on a Friday night and didn't like what the algorithm served him. So he built his own. In about 𝟮 𝗵𝗼𝘂𝗿𝘀 he made HackerTube: 𝟲,𝟰𝟱𝟰 𝘃𝗶𝗱𝗲𝗼𝘀 pulled from 𝟰𝟬𝗞 𝗛𝗮𝗰𝗸𝗲𝗿 𝗡𝗲𝘄𝘀 𝗰𝗼𝗺𝗺𝗲𝗻𝘁𝘀 that recommend a YouTube video, from 2007 to 2026, sorted into 𝟭𝟱𝟬 𝘁𝗼𝗽𝗶𝗰𝘀. He also published the data, with a note that it's there to "𝘨𝘪𝘷𝘦 𝘵𝘰 𝘺𝘰𝘶𝘳 𝘢𝘨𝘦𝘯𝘵 𝘵𝘰 𝘦𝘢𝘴𝘪𝘭𝘺 𝘤𝘳𝘦𝘢𝘵𝘦 𝘺𝘰𝘶𝘳 𝘰𝘸𝘯 𝘷𝘦𝘳𝘴𝘪𝘰𝘯." 𝗦𝗼 𝗜 𝗱𝗶𝗱. 𝗔𝗻𝗱 𝗜 𝘄𝗲𝗻𝘁 𝗼𝗻𝗲 𝘀𝘁𝗲𝗽 𝗳𝘂𝗿𝘁𝗵𝗲𝗿. HackerTube shows what engineers keep recommending. YouTube knows how many people pressed play. 𝗡𝗲𝗿𝗱𝗟𝗼𝗿𝗲 𝗽𝘂𝘁𝘀 𝘁𝗵𝗲 𝘁𝘄𝗼 𝘀𝗶𝗱𝗲 𝗯𝘆 𝘀𝗶𝗱𝗲, 𝘃𝗶𝗱𝗲𝗼 𝗯𝘆 𝘃𝗶𝗱𝗲𝗼, 𝘀𝗼 𝘆𝗼𝘂 𝗰𝗮𝗻 𝘀𝗲𝗲 𝘄𝗵𝗲𝗿𝗲 𝘁𝗵𝗲𝘆 𝗱𝗶𝘀𝗮𝗴𝗿𝗲𝗲. I've open-sourced it under MIT. 𝗪𝗵𝗮𝘁 𝗜𝘁 𝗦𝗵𝗼𝘄𝘀 • 𝗦𝗮𝗺𝗲 𝗽𝗿𝗮𝗶𝘀𝗲, 𝗱𝗶𝗳𝗳𝗲𝗿𝗲𝗻𝘁 𝗿𝗲𝗮𝗰𝗵: two videos with exactly the same number of HN recommenders. On 3 Nov: 16 each, 3Blue1Brown's neural network intro at 𝟮𝟱,𝟬𝟴𝟴,𝟮𝟲𝟱 𝘃𝗶𝗲𝘄𝘀, Freya Holmér's "The Beauty of Bézier Curves" at 𝟮,𝟮𝟯𝟭,𝟴𝟭𝟳. • 𝗣𝗼𝗽𝘂𝗹𝗮𝗿 𝘃𝘀 𝗿𝗲𝗺𝗲𝗺𝗯𝗲𝗿𝗲𝗱: every video on one map, in four zones (hidden gems, shared classics, crowd hits, quiet corner). You set the lines. • 𝗛𝗶𝗱𝗱𝗲𝗻 𝗴𝗲𝗺𝘀: loved by HN, watched by few, each with a real HN quote. • 𝗔𝗹𝘀𝗼: the channels HN trusts most, the best picks by year, and HN's running video jokes. 𝗧𝗿𝘆 𝗜𝘁 𝗪𝗶𝘁𝗵𝗼𝘂𝘁 𝗮 𝗞𝗲𝘆 There's a 𝗱𝗲𝗺𝗼 𝗺𝗼𝗱𝗲: one click, no API key, no download. It runs on a small built-in sample of 𝟲𝟮𝟴 𝘃𝗶𝗱𝗲𝗼𝘀 with view counts that are 𝗺𝗮𝗱𝗲 𝘂𝗽, and every page says so. For real numbers you bring your own free YouTube API key. A run costs about 𝟭 𝗾𝘂𝗼𝘁𝗮 𝘂𝗻𝗶𝘁 𝗽𝗲𝗿 𝟱𝟬 𝘃𝗶𝗱𝗲𝗼𝘀, and the free daily quota is 𝟭𝟬,𝟬𝟬𝟬. 𝗙𝗼𝗿 𝗕𝘂𝗶𝗹𝗱𝗲𝗿𝘀 • 𝗧𝘄𝗼 𝗣𝘆𝘁𝗵𝗼𝗻 𝗱𝗲𝗽𝗲𝗻𝗱𝗲𝗻𝗰𝗶𝗲𝘀: Streamlit and pandas. That's the whole list. • The interactive pages are plain HTML, CSS and canvas. No chart libraries. • It runs on your own computer. No accounts, no tracking, and your key stays on your machine. • 𝟯𝟬 𝘁𝗲𝘀𝘁𝘀, run on Python 3.10 and 3.13. 𝗧𝗵𝗲 𝗛𝗼𝗻𝗲𝘀𝘁 𝗣𝗮𝗿𝘁 • View counts are a snapshot, shown exactly as YouTube returns them. The app never turns them into scores. • HackerTube's labels were written by an LLM, so expect 𝘀𝗼𝗺𝗲 𝗻𝗼𝗶𝘀𝗲. • A talk can look unwatched because another upload holds the views. There's a check for that, but it costs quota. • It's 𝗱𝗮𝘆𝘀 𝗼𝗹𝗱. Expect rough edges. 𝘔𝘢𝘺𝘣𝘦 𝘵𝘩𝘦 𝘣𝘦𝘴𝘵 𝘳𝘦𝘤𝘰𝘮𝘮𝘦𝘯𝘥𝘢𝘵𝘪𝘰𝘯 𝘦𝘯𝘨𝘪𝘯𝘦 𝘪𝘴 𝘢 𝘧𝘦𝘸 𝘵𝘩𝘰𝘶𝘴𝘢𝘯𝘥 𝘦𝘯𝘨𝘪𝘯𝘦𝘦𝘳𝘴 𝘵𝘦𝘭𝘭𝘪𝘯𝘨 𝘦𝘢𝘤𝘩 𝘰𝘵𝘩𝘦𝘳 𝘸𝘩𝘢𝘵 𝘵𝘰 𝘸𝘢𝘵𝘤𝘩. Links in reply 👇 NerdLore (open source, MIT): github.com/saeedkhattak/nerd…

Not a New Voice Agent, a New Way to Listen: Soniox STT v6 You call an airline bot on speakerphone. It says: "I can rebook you through Porto or Faro." You say: "Faro." But the bot's own voice is still coming out of your speaker and into your mic. So the bot hears itself, thinks you interrupted, and stops talking. Or it mishears a one-word answer. 𝗧𝗵𝗮𝘁 𝗶𝘀 𝘁𝗵𝗲 𝗽𝗿𝗼𝗯𝗹𝗲𝗺 𝗦𝗼𝗻𝗶𝗼𝘅 𝗦𝗧𝗧 𝘃𝟲 𝗴𝗼𝗲𝘀 𝗮𝗳𝘁𝗲𝗿. 𝗪𝗵𝘆 𝗜𝘁 𝗛𝗮𝗽𝗽𝗲𝗻𝘀 A voice agent is usually three models in a row: speech-to-text, an LLM, text-to-speech. The odd thing is that the first model has usually been 𝗵𝗮𝗹𝗳-𝗱𝗲𝗮𝗳 𝘁𝗼 𝘁𝗵𝗲 𝗰𝗼𝗻𝘃𝗲𝗿𝘀𝗮𝘁𝗶𝗼𝗻. It hears the caller but has no idea what the agent just asked. So when the caller says "8", the STT model has to guess: eight? ate? a stray noise? 𝗪𝗵𝗮𝘁'𝘀 𝗡𝗲𝘄 Soniox released STT v6 on 3 November. It comes in two versions: 𝚜𝚝𝚝-𝚛𝚝-𝚟𝟼 for live audio and 𝚜𝚝𝚝-𝚊𝚜𝚢𝚗𝚌-𝚟𝟼 for recordings. The headline is two features for voice agents: • 𝗔𝗴𝗲𝗻𝘁 𝘁𝗲𝘅𝘁 𝗶𝗻𝗽𝘂𝘁: you stream the agent's reply text to the speech model, as you send it to text-to-speech. Now it knows the last question was "7 or 8 p.m.?", so a short answer like "8" or "Faro" is easier to get right. • 𝗔𝗴𝗲𝗻𝘁 𝘃𝗼𝗶𝗰𝗲 𝗳𝗶𝗹𝘁𝗲𝗿𝗶𝗻𝗴: the second trick follows from the first. If you know what the agent is saying, 𝘆𝗼𝘂 𝗰𝗮𝗻 𝘀𝘂𝗯𝘁𝗿𝗮𝗰𝘁 𝗶𝘁. v6 uses that text to filter the agent's voice out of the mic audio, and keeps listening for real interruptions. Soniox says this means 𝗳𝗲𝘄𝗲𝗿 𝗳𝗮𝗹𝘀𝗲 𝗶𝗻𝘁𝗲𝗿𝗿𝘂𝗽𝘁𝗶𝗼𝗻𝘀 on speakerphones, laptops and in cars. 𝗔𝗻 𝗔𝗻𝗮𝗹𝗼𝗴𝘆 It's like noise-cancelling headphones that know the song in advance. 𝘊𝘢𝘯𝘤𝘦𝘭𝘭𝘪𝘯𝘨 𝘢 𝘴𝘰𝘶𝘯𝘥 𝘪𝘴 𝘮𝘶𝘤𝘩 𝘦𝘢𝘴𝘪𝘦𝘳 𝘸𝘩𝘦𝘯 𝘺𝘰𝘶 𝘢𝘭𝘳𝘦𝘢𝘥𝘺 𝘬𝘯𝘰𝘸 𝘸𝘩𝘢𝘵 𝘪𝘵 𝘪𝘴. 𝗔𝗹𝘀𝗼 𝗜𝗻 𝘃𝟲 • 𝗩𝗼𝗶𝗰𝗲 𝘁𝘆𝗽𝗶𝗻𝗴: better numbers, dates, currencies, IDs and names, in 60+ languages. • 𝗠𝗲𝗲𝘁𝗶𝗻𝗴𝘀: better "who said what", including overlapping voices. • 𝗟𝗮𝗻𝗴𝘂𝗮𝗴𝗲𝘀: better accents and mid-sentence language switching. 𝗙𝗼𝗿 𝗕𝘂𝗶𝗹𝗱𝗲𝗿𝘀 • 𝗨𝗽𝗴𝗿𝗮𝗱𝗲: change the model name. v5 is removed on 𝟭𝟱 𝗗𝗲𝗰𝗲𝗺𝗯𝗲𝗿 𝟮𝟬𝟮𝟲. • 𝗣𝗿𝗶𝗰𝗲: $𝟬.𝟭𝟮/𝗵𝗼𝘂𝗿 live, $𝟬.𝟭𝟬/𝗵𝗼𝘂𝗿 for files. • 𝗟𝗶𝗺𝗶𝘁𝘀: the two voice-agent features work in live mode only, and the docs say to keep normal echo cancellation on. v6 cleans up what it misses. 𝗪𝗵𝗮𝘁 𝗜𝘁 𝗕𝘂𝗶𝗹𝗱𝘀 𝗢𝗻 Voice apps already use acoustic echo cancellation in the call layer, usually WebRTC's. Deepgram Flux moved turn detection into the speech model. End-to-end voice models like OpenAI's Realtime API hear both sides because one model does everything. 𝗪𝗵𝘆 𝗧𝗵𝗲 𝗗𝗲𝘀𝗶𝗴𝗻 𝗠𝗮𝘁𝘁𝗲𝗿𝘀 v6 doesn't replace the pipeline. You still pick your own LLM and TTS. It just 𝘀𝘁𝗼𝗽𝘀 𝘁𝗵𝗿𝗼𝘄𝗶𝗻𝗴 𝗮𝘄𝗮𝘆 𝗶𝗻𝗳𝗼𝗿𝗺𝗮𝘁𝗶𝗼𝗻 𝘁𝗵𝗲 𝘀𝘆𝘀𝘁𝗲𝗺 𝗮𝗹𝗿𝗲𝗮𝗱𝘆 𝗵𝗮𝗱. 𝗧𝗵𝗲 𝗛𝗼𝗻𝗲𝘀𝘁 𝗣𝗮𝗿𝘁 • 𝗡𝗼 𝘃𝟲 𝗻𝘂𝗺𝗯𝗲𝗿𝘀: the launch post gives no accuracy or latency figures. Soniox's benchmark page still shows v5. • 𝗖𝗹𝗼𝘀𝗲 𝗿𝗮𝗰𝗲: on Pipecat's open benchmark, v5 scored 𝟭.𝟭𝟭% semantic word error rate against Azure's 𝟬.𝟵𝟴%. The benchmark says gaps under 𝗮𝗯𝗼𝘂𝘁 𝟬.𝟮𝟱 𝗽𝗼𝗶𝗻𝘁𝘀 can't reliably rank providers. • 𝗪𝗵𝗲𝗿𝗲 𝘃𝟱 𝗯𝗲𝗮𝘁 𝗔𝘇𝘂𝗿𝗲: speed. 𝟮𝟲𝟬 𝗺𝘀 median to the final transcript, against 𝟭,𝟬𝟭𝟲 𝗺𝘀 for Azure. • "𝗪𝗼𝗿𝗹𝗱'𝘀 𝗳𝗶𝗿𝘀𝘁": that is Soniox's own claim. 𝗧𝗵𝗲 𝗯𝗶𝗴𝗴𝗲𝗿 𝗽𝗼𝗶𝗻𝘁: 𝘷𝘰𝘪𝘤𝘦 𝘢𝘨𝘦𝘯𝘵𝘴 𝘧𝘢𝘪𝘭 𝘭𝘦𝘴𝘴 𝘧𝘳𝘰𝘮 𝘣𝘢𝘥 𝘩𝘦𝘢𝘳𝘪𝘯𝘨 𝘵𝘩𝘢𝘯 𝘧𝘳𝘰𝘮 𝘯𝘰𝘵 𝘬𝘯𝘰𝘸𝘪𝘯𝘨 𝘸𝘩𝘢𝘵 𝘵𝘩𝘦𝘺 𝘫𝘶𝘴𝘵 𝘴𝘢𝘪𝘥. Links in reply 👇 v6 docs (agent text input, echo filtering): Soniox STT v6 announcement: soniox.com/blog/soniox-stt-v…

𝗧𝗵𝗲 𝗙𝘂𝗹𝗹 𝗦𝘁𝗼𝗿𝘆 𝗢𝗳 𝗧𝗵𝗲 $𝟲.𝟯𝗕 𝗕𝗶𝗻𝗮𝗻𝗰𝗲 𝗛𝗮𝗰𝗸, 𝗙𝗿𝗼𝗺 𝗔 𝗝𝗼𝗯 𝗢𝗳𝗳𝗲𝗿 𝗧𝗼 𝗘𝗺𝗽𝘁𝘆 𝗪𝗮𝗹𝗹𝗲𝘁𝘀 The biggest crypto theft in history started with a job interview. A software engineer at Binance, the world’s largest crypto exchange, had a profile on a job site. It mentioned they worked at Binance. One day a message came in from a “Project Director” at another company. There was a role. Interested? First came an interview with someone from HR. Then a second round, with the company’s CEO. None of it was real. Not the HR person, not the CEO, not the company. On the other side were North Korean hackers, and they were about to take $6.3 billion. Here’s how they did it, based on what Binance has now made public. After the interviews, the “company” sent a take-home assignment. That’s normal in tech hiring. You download a small task, work on it, send it back. The engineer downloaded it. Hidden inside was a way into the laptop. From then on, the hackers could use that engineer’s work computer as if they were sitting at it. But the laptop was just the front door. What they really wanted was Binance’s wallets. In crypto, a wallet is where coins are kept. Whoever controls it can send those coins anywhere, and once they’re sent, there’s no bank to call and no “undo” button. Binance holds billions of dollars of its customers’ coins in its wallets. So the hackers went after the engineers who build and run them. They already knew who those people were, because a second trap had been running at the same time. The hackers ran an online ad pushing a fake version of Homebrew, a free tool developers use to install software on their computers. Another Binance engineer clicked the ad and installed the fake one. Binance caught it and removed it. Too late. The hackers had already learned that engineer’s role, and who else at Binance managed the wallets. So now they had a laptop inside Binance and a list of names. Next, they turned Binance’s own tools against it. Binance gives staff an internal AI tool for their work. From the hacked laptop, the attackers used it to take over two more engineers: one who runs the wallet systems day to day, and one who writes wallet software. Binance hasn’t said exactly how. On one of those computers they found a “container deployment permission token.” In plain English: big companies run their software inside sealed boxes called containers. A deployment token is a digital pass that lets you launch or change what runs in those boxes. Binance says the attacker got into one of these containers, a working part of Binance’s own system, and from there sent out withdrawals Binance never approved. That’s the part that should scare every company. They never stole the wallets’ private keys, the master passwords to crypto wallets. Binance says there’s no sign any key leaked. They didn’t need one. They got into Binance’s own machinery and had it send the money out. On Thursday, October 29, they took over a series of Binance wallets and drained them. $6.3 billion. More than four times the old record, the roughly $1.5 billion stolen from Bybit in 2025, also blamed on North Korea. Binance calls it a “monthslong scheme.” That means all of this happened before a single coin moved: a fake company, a fake HR person, a fake CEO doing a second-round interview, a fake download pushed through ads, and quietly mapping out who did what inside Binance. Some of the money was saved. Part of what was taken was stablecoins, crypto tokens pegged 1:1 to the US dollar. Unlike Bitcoin, they’re issued by companies, and those companies can freeze them. Tether and Circle froze the stolen ones, and about $2.5 billion has been recovered. Binance shut withdrawals for about 48 hours. It says it has refilled 98% of the stolen coins from its own reserves and that customer funds are safe. Two details stay with me. Binance has a policy that employees in roles that aren’t public-facing shouldn’t have LinkedIn profiles. It had already seen hackers targeting its staff on LinkedIn, and it says it stopped 57 North Korea-linked hacking attempts this year. This one it didn’t stop. And none of it needed a genius hack to get started. It needed a believable job offer and one engineer willing to do a take-home test. Links in First Reply 👇👇 “This should not have happened,” Binance said. thehackernews.com/2026/10/bi…

𝗔𝗹𝗹 𝗬𝗼𝘂 𝗡𝗲𝗲𝗱 𝘁𝗼 𝗞𝗻𝗼𝘄 𝗔𝗯𝗼𝘂𝘁 𝗖𝗹𝗮𝘂𝗱𝗲'𝘀 𝗡𝗲𝘄 𝗠𝗲𝘀𝘀𝗮𝗴𝗲 𝗧𝗵𝗿𝗲𝗮𝗱𝘀 Picture a coding agent two hours into a job. It has opened 40 files, run the tests a dozen times and written a long plan. Now it wants to take one tiny next step. Until now, to do that, it had to send Anthropic's API 𝘁𝗵𝗲 𝗲𝗻𝘁𝗶𝗿𝗲 𝗰𝗼𝗻𝘃𝗲𝗿𝘀𝗮𝘁𝗶𝗼𝗻 𝗮𝗴𝗮𝗶𝗻. Every file, every test log, every earlier message. Then again on the next step. And the next. 𝗢𝗻 𝗡𝗼𝘃𝗲𝗺𝗯𝗲𝗿 𝟮, 𝗔𝗻𝘁𝗵𝗿𝗼𝗽𝗶𝗰 𝗮𝗱𝗱𝗲𝗱 "𝗺𝗲𝘀𝘀𝗮𝗴𝗲 𝘁𝗵𝗿𝗲𝗮𝗱𝘀", 𝗮𝗻𝗱 𝘁𝗵𝗮𝘁 𝗰𝗵𝗮𝗻𝗴𝗲𝘀. Here's the difference, with made-up sizes just to show the idea: ❌ 𝗕𝗲𝗳𝗼𝗿𝗲: 𝗻𝗼 𝘁𝗵𝗿𝗲𝗮𝗱 Turn 1: send the whole conversation (800 KB) Turn 2: send all of it again, plus your new message (802 KB) Turn 3: all of it again, plus the next message (804 KB) ✅ 𝗔𝗳𝘁𝗲𝗿: 𝘄𝗶𝘁𝗵 𝗮 𝘁𝗵𝗿𝗲𝗮𝗱 Turn 1: send the whole conversation once (800 KB) Turn 2: send only your new message and the ID of Claude's last reply (about 2 KB) Turn 3: only the next message and the latest ID (about 2 KB) Anthropic's servers stitch the full conversation back together before Claude reads it. So Claude still sees everything. Your app just stops carrying it back and forth. Those sizes aren't far from real life. Anthropic says the typical request in long Claude Code sessions went from 𝗮𝗯𝗼𝘂𝘁 𝟴𝟬𝟬 𝗞𝗕 𝘁𝗼 𝟭.𝟴 𝗞𝗕. 𝗦𝗼 𝘄𝗵𝗮𝘁 𝗵𝗮𝗽𝗽𝗲𝗻𝘀 𝘁𝗼 𝘁𝗶𝗺𝗲 𝗮𝗻𝗱 𝗽𝗿𝗶𝗰𝗲? ⏱️ 𝗧𝗶𝗺𝗲 𝗴𝗼𝗲𝘀 𝗱𝗼𝘄𝗻. Without a thread, the API has to receive, check and tokenize the whole conversation again before Claude can start. With a thread, most of that work covers only the new part. On really big sessions (500K to 1M tokens), Anthropic says Claude started replying 𝗮𝗯𝗼𝘂𝘁 𝟯𝟱% 𝘀𝗼𝗼𝗻𝗲𝗿. 💰 𝗣𝗿𝗶𝗰𝗲 𝘀𝘁𝗮𝘆𝘀 𝘁𝗵𝗲 𝘀𝗮𝗺𝗲. Claude still reads the full conversation on every turn, so you pay for the same tokens as before. Rate limits don't change either. If you want to cut cost, that's still the job of prompt caching. People get confused about who this affects, so let's split it in two. 𝗜𝗳 𝘆𝗼𝘂 𝗷𝘂𝘀𝘁 𝘂𝘀𝗲 𝗖𝗹𝗮𝘂𝗱𝗲 𝗖𝗼𝗱𝗲 You don't have to do anything. Anthropic says threads are already 𝗼𝗻 𝗯𝘆 𝗱𝗲𝗳𝗮𝘂𝗹𝘁 𝗳𝗼𝗿 𝗺𝗼𝘀𝘁 𝗖𝗹𝗮𝘂𝗱𝗲 𝗖𝗼𝗱𝗲 𝘂𝘀𝗲𝗿𝘀. It all happens in the background. There's no new button, no setting and nothing on screen. You chat and code the way you always have. In long sessions, replies may just start a bit sooner. If you step away for a while, the thread expires 𝟱 𝗺𝗶𝗻𝘂𝘁𝗲𝘀 after the last request. The full history then gets sent again to start a new one. You won't see that happen. That one reply just won't get the speed boost. Anthropic has only talked about Claude Code here. It hasn't said anything about the regular Claude chat app. 𝗜𝗳 𝘆𝗼𝘂 𝗯𝘂𝗶𝗹𝗱 𝗼𝗻 𝘁𝗵𝗲 𝗔𝗣𝗜 Here it's the opposite: 𝗻𝗼𝘁𝗵𝗶𝗻𝗴 𝗰𝗵𝗮𝗻𝗴𝗲𝘀 𝘂𝗻𝗹𝗲𝘀𝘀 𝘆𝗼𝘂 𝘁𝘂𝗿𝗻 𝗶𝘁 𝗼𝗻. The easiest way is the thread helper in Anthropic's SDK. In Python it looks like this: client = anthropic.Anthropic() thread = client.beta.messages.thread( model="claude-opus-5-5", max_tokens=1024, system="You are a coding assistant.") thread.send({"role": "user", "content": "Fix the failing test."}) thread.send({"role": "user", "content": "Now run it again."}) The first send() sends the full conversation and starts the thread. Every send() after that sends only the new message. The helper adds the beta header for you, and if a thread has expired, it sends the full history again on its own. A few things worth knowing before you switch it on: If your chats are only a few thousand tokens long, you probably won't notice much. This is built for long-running agents. 𝗢𝘁𝗵𝗲𝗿𝘀 𝗴𝗼𝘁 𝗵𝗲𝗿𝗲 𝗳𝗶𝗿𝘀𝘁, 𝗶𝗻 𝗮 𝗱𝗶𝗳𝗳𝗲𝗿𝗲𝗻𝘁 𝘄𝗮𝘆 OpenAI's Responses API and Google's Interactions API can already continue a conversation from an ID. 𝗙𝗼𝗿 𝗹𝗼𝗻𝗴-𝗿𝘂𝗻𝗻𝗶𝗻𝗴 𝗮𝗴𝗲𝗻𝘁𝘀, 𝗷𝘂𝘀𝘁 𝗺𝗼𝘃𝗶𝗻𝗴 𝘁𝗵𝗲 𝗰𝗼𝗻𝘃𝗲𝗿𝘀𝗮𝘁𝗶𝗼𝗻 𝗮𝗿𝗼𝘂𝗻𝗱 𝗵𝗮𝘀 𝗯𝗲𝗰𝗼𝗺𝗲 𝗽𝗮𝗿𝘁 𝗼𝗳 𝘁𝗵𝗲 𝘄𝗮𝗶𝘁. claude.com/resources/article…

Saw this and wondered why Elon still works like this when he doesn't need to. I got the answer pretty quickly. Tbh a yacht is fun for a week. After that it's the same sea every morning. Some might disagree , but that is fine . Most of us think the point of money is to stop working. But rest only feels good when there's work around it. Take the work away and relaxing becomes its own kind of boredom. What doesn't get boring is building something nobody has built before. You never really finish it, so it keeps you hooked for years. If Elon had cashed out and moved onto a boat, he wouldn't be Elon, and none of us would be talking about his 3am office nights. People with lots of money and nothing to build often drift into bad habits. People deep in building something rarely have time for them. On a much smaller scale, my best late nights are the ones spent building something. It feels more like rest than any day off.

𝗡𝘃𝗶𝗱𝗶𝗮'𝘀 𝟱𝟭% 𝗥𝘂𝗹𝗲: 𝗪𝗵𝗮𝘁 𝗜𝘁 𝗚𝗶𝘃𝗲𝘀 𝗬𝗼𝘂 𝗔𝗻𝗱 𝗪𝗵𝗮𝘁 𝗜𝘁 𝗗𝗼𝗲𝘀𝗻'𝘁 Picture a data centre in Singapore. Racks full of Nvidia chips. Its biggest clients are Chinese AI companies that can't buy those chips at home, so they rent them. 𝗧𝗵𝗲 𝗨𝗻𝗱𝗲𝗿𝗹𝘆𝗶𝗻𝗴 𝗜𝗱𝗲𝗮 Compute is a physical thing, but access to it isn't. You can stop a box at a border. It's much harder to stop someone logging in to a box in another country. So there are two doors: • 𝗕𝘂𝘆𝗶𝗻𝗴: export controls, plus Nvidia's "white list", which per the FT more than halved approved Asian buyers this year. • 𝗥𝗲𝗻𝘁𝗶𝗻𝗴: Chinese companies leasing GPUs in overseas data centres, including for model training. Close the first door and traffic moves to the second. 𝗧𝗵𝗶𝘀 𝗿𝘂𝗹𝗲 𝗶𝘀 𝗮 𝘃𝗮𝗹𝘃𝗲 𝗼𝗻 𝘁𝗵𝗲 𝘀𝗲𝗰𝗼𝗻𝗱 𝗱𝗼𝗼𝗿. 𝗔𝗻 𝗔𝗻𝗮𝗹𝗼𝗴𝘆 It's a bit like a landlord telling a building manager: no more than 49% of tenants from one place. Not a law. A lease condition, enforced by whoever signs the lease. 𝗡𝘃𝗶𝗱𝗶𝗮 𝗻𝗼𝘄 𝘄𝗮𝗻𝘁𝘀 𝘁𝗵𝗮𝘁 𝘁𝗼 𝗰𝗵𝗮𝗻𝗴𝗲. 𝗪𝗵𝗮𝘁 𝗛𝗮𝗽𝗽𝗲𝗻𝗲𝗱 According to the Financial Times (five people familiar with the matter), Nvidia has told cloud providers that 𝗮𝘁 𝗹𝗲𝗮𝘀𝘁 𝟱𝟭% of their services must go to non-Chinese companies. There was no such limit before. • 𝗡𝗼 𝗽𝗮𝗽𝗲𝗿𝘄𝗼𝗿𝗸: the instruction was given verbally, through Nvidia's sales and regional teams. • 𝗪𝗼𝗿𝗹𝗱𝘄𝗶𝗱𝗲: it applies to clients globally, not only in Asia (two people told the FT). • 𝗡𝗲𝘄 𝗱𝗲𝗮𝗹𝘀 𝗼𝗻𝗹𝘆: existing leases are not affected. • 𝗦𝗶𝗻𝗰𝗲 𝘁𝗵𝗲 𝘀𝘂𝗺𝗺𝗲𝗿: two people said Nvidia did it pre-emptively, to cut criticism from US lawmakers. 𝗪𝗵𝘆 𝗥𝗲𝗻𝘁𝗶𝗻𝗴 𝗠𝗮𝘁𝘁𝗲𝗿𝘀 US export rules mostly control where chips are shipped. Renting them in a data centre abroad has been the gap. Many Chinese groups used overseas leases for heavy AI work, including model training. Nvidia had already tightened the front door. This year it built a "white list" of approved Asian buyers and 𝗺𝗼𝗿𝗲 𝘁𝗵𝗮𝗻 𝗵𝗮𝗹𝘃𝗲𝗱 it, per the FT. That pushed more Chinese firms toward renting. This new rule goes after 𝘁𝗵𝗲 𝗿𝗲𝗻𝘁𝗶𝗻𝗴. 𝗪𝗵𝗮𝘁 𝗜𝘁 𝗗𝗼𝗲𝘀 𝗧𝗼 𝗣𝗿𝗶𝗰𝗲𝘀 A rack of B300 chips sells for 𝗮𝗯𝗼𝘂𝘁 $𝟲𝟬𝟬,𝟬𝟬𝟬 internationally. One person told the FT a rack fetched 𝗥𝗺𝗯𝟭𝟴.𝟱𝗺𝗻 (𝗮𝗯𝗼𝘂𝘁 $𝟮.𝟴𝗺𝗻) in China last month. That is 𝗿𝗼𝘂𝗴𝗵𝗹𝘆 𝟰.𝟳 𝘁𝗶𝗺𝗲𝘀 the price. The B300 can't legally be sold in China. 𝗙𝗼𝗿 𝗖𝗹𝗼𝘂𝗱 𝗣𝗿𝗼𝘃𝗶𝗱𝗲𝗿𝘀 Chinese tech groups are the biggest clients for data centres in places like south-east Asia. A Singapore-based provider told the FT it is hard to find enough non-Chinese clients outside the US. Providers that can't find them may have to 𝘀𝗵𝗿𝗶𝗻𝗸. New contracts only. 𝗧𝗵𝗲 𝗠𝗮𝗽 𝗢𝗳 𝗖𝗵𝗶𝗻𝗮 𝗖𝗼𝗺𝗽𝘂𝘁𝗲 𝗖𝗼𝗻𝘁𝗿𝗼𝗹𝘀, 𝗢𝗻𝗲 𝗟𝗶𝗻𝗲 𝗘𝗮𝗰𝗵 • 𝗨𝗦 𝗲𝘅𝗽𝗼𝗿𝘁 𝗰𝗼𝗻𝘁𝗿𝗼𝗹𝘀: block shipping top chips like the B300 to China. • 𝗕𝗜𝗦 𝗿𝘂𝗹𝗲 (𝗝𝗮𝗻𝘂𝗮𝗿𝘆): license applicants must list remote cloud users in China and other listed countries. • 𝗕𝗜𝗦 𝗴𝘂𝗶𝗱𝗮𝗻𝗰𝗲 (𝗠𝗮𝘆 𝟯𝟭): a license is needed for advanced chips going to companies headquartered in China or Macau, wherever they are. • 𝗡𝘃𝗶𝗱𝗶𝗮'𝘀 𝘄𝗵𝗶𝘁𝗲 𝗹𝗶𝘀𝘁 (𝘁𝗵𝗶𝘀 𝘆𝗲𝗮𝗿): approved Asian buyers more than halved, per the FT. • 𝗥𝗲𝗺𝗼𝘁𝗲 𝗔𝗰𝗰𝗲𝘀𝘀 𝗦𝗲𝗰𝘂𝗿𝗶𝘁𝘆 𝗔𝗰𝘁: passed the House 369-22 in January; sitting with the Senate Banking Committee. • 𝗡𝘃𝗶𝗱𝗶𝗮'𝘀 𝟱𝟭% 𝗿𝘂𝗹𝗲 (𝗻𝗼𝘄): aimed at the cloud provider's customer mix, not the chip. 𝗪𝗵𝗲𝗿𝗲 𝗧𝗵𝗶𝘀 𝗙𝗶𝘁𝘀 Government rules set the floor. Nvidia is adding its own layer on top, so lawmakers have less to criticise (two people told the FT that was the motive). 𝗪𝗵𝗼 𝗙𝗲𝗲𝗹𝘀 𝗜𝘁 • 𝗖𝗹𝗼𝘂𝗱 𝗽𝗿𝗼𝘃𝗶𝗱𝗲𝗿𝘀 𝗶𝗻 𝘀𝗼𝘂𝘁𝗵-𝗲𝗮𝘀𝘁 𝗔𝘀𝗶𝗮: Chinese tech groups are their biggest clients. • 𝗖𝗵𝗶𝗻𝗲𝘀𝗲 𝗔𝗜 𝗰𝗼𝗺𝗽𝗮𝗻𝗶𝗲𝘀: already paying a premium. A B300 rack, 𝗮𝗯𝗼𝘂𝘁 $𝟲𝟬𝟬𝗸 abroad, reportedly fetched $𝟮.𝟴𝗺𝗻 in China last month (one deal). 𝗧𝗵𝗲 𝗯𝗶𝗴𝗴𝗲𝗿 𝗽𝗼𝗶𝗻𝘁: 𝘵𝘩𝘦 𝘤𝘩𝘪𝘱 𝘸𝘢𝘳 𝘪𝘴 𝘮𝘰𝘷𝘪𝘯𝘨 𝘧𝘳𝘰𝘮 𝘸𝘩𝘰 𝘤𝘢𝘯 𝘣𝘶𝘺 𝘤𝘩𝘪𝘱𝘴 𝘵𝘰 𝘸𝘩𝘰 𝘤𝘢𝘯 𝘳𝘦𝘯𝘵 𝘵𝘩𝘦𝘮. FT (original report): ft.com/content/1902df70-1380… Links in reply 👇

Who is this reply even for? Saw a reply today under my NVIDIA auto-ontology repo post today . It just said "@bot save this". I assumed it was a joke, so I looked it up. @bot is xAI's Grok Bot, an agent app they launched in August 2026. You tag it under a post and the post goes to your own agent, which runs on its own computer in the cloud all day. So that reply wasn't really for the person who shared the repo. It was for someone's private assistant, left out in public where the rest of us could read it. Feels a bit like leaving a note for your cleaner on a stranger's front door. I get why people do it though. I have hundreds of bookmarks I've never opened. Every time I save something I tell myself I'll read it later, and I don't. At least an agent might actually go through the list. I read that repo properly because I'm building an ontology layer on top of a lineage graph at work. I doubt most people saving it will open the README. Their bots probably will. Still can't decide how I feel about that.