Gul Saeed Khattak

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.

𝟯 𝗢𝗽𝗲𝗻-𝗦𝗼𝘂𝗿𝗰𝗲 𝗧𝗼𝗼𝗹𝘀 𝗥𝗮𝗰𝗶𝗻𝗴 𝘁𝗼 𝗠𝗮𝗻𝗮𝗴𝗲 𝗬𝗼𝘂𝗿 𝗔𝗜 𝗔𝗴𝗲𝗻𝘁𝘀 You have Claude Code in one terminal. Codex in another. Five more agents running somewhere. 𝗪𝗵𝗼'𝘀 𝗺𝗮𝗻𝗮𝗴𝗶𝗻𝗴 𝘁𝗵𝗲𝗺? Three open-source projects want that job. Same goal, very different ideas. 𝗣𝗮𝗽𝗲𝗿𝗰𝗹𝗶𝗽: 𝘁𝗵𝗲 𝗰𝗼𝗺𝗽𝗮𝗻𝘆 (𝟭𝟭𝟬𝗸 ⭐) "𝘐𝘧 𝘖𝘱𝘦𝘯𝘊𝘭𝘢𝘸 𝘪𝘴 𝘢𝘯 𝘦𝘮𝘱𝘭𝘰𝘺𝘦𝘦, 𝘗𝘢𝘱𝘦𝘳𝘤𝘭𝘪𝘱 𝘪𝘴 𝘵𝘩𝘦 𝘤𝘰𝘮𝘱𝘢𝘯𝘺." • Hire agents into roles: CEO, engineer, marketer • Set goals and budgets, approve plans • Not just code: research, content, operations too 𝗢𝗽𝗲𝗻𝗛𝗮𝗻𝗱𝘀: 𝘁𝗵𝗲 𝗲𝗻𝗴𝗶𝗻𝗲𝗲𝗿𝗶𝗻𝗴 𝘁𝗲𝗮𝗺 (𝟵𝟮𝗸 ⭐) Now leads with Agent Canvas, a "𝘥𝘦𝘷𝘦𝘭𝘰𝘱𝘦𝘳 𝘤𝘰𝘯𝘵𝘳𝘰𝘭 𝘤𝘦𝘯𝘵𝘦𝘳 𝘧𝘰𝘳 𝘤𝘰𝘥𝘪𝘯𝘨 𝘢𝘨𝘦𝘯𝘵𝘴." • The only one of the three with 𝗶𝘁𝘀 𝗼𝘄𝗻 𝗰𝗼𝗱𝗶𝗻𝗴 𝗮𝗴𝗲𝗻𝘁 • Also runs Claude Code, Codex, Gemini and others • Automations with Slack, GitHub and Linear. Self-host or use their cloud 𝗥𝘂𝗳𝗹𝗼: 𝘁𝗵𝗲 𝗽𝗼𝘄𝗲𝗿 𝘁𝗼𝗼𝗹 (𝟴𝟭𝗸 ⭐) "𝘈𝘯 𝘢𝘨𝘦𝘯𝘵 𝘮𝘦𝘵𝘢-𝘩𝘢𝘳𝘯𝘦𝘴𝘴 𝘧𝘰𝘳 𝘊𝘭𝘢𝘶𝘥𝘦 𝘊𝘰𝘥𝘦 𝘢𝘯𝘥 𝘊𝘰𝘥𝘦𝘹." • Lives inside Claude Code as plugins • Runs "swarms" of agents, with routing and policy rules • Built for developers who stay in the terminal 𝗧𝗵𝗲 𝘀𝗶𝗺𝗽𝗹𝗲 𝘃𝗲𝗿𝘀𝗶𝗼𝗻: • Paperclip manages 𝘁𝗵𝗲 𝗰𝗼𝗺𝗽𝗮𝗻𝘆 • OpenHands runs 𝘁𝗵𝗲 𝗲𝗻𝗴𝗶𝗻𝗲𝗲𝗿𝗶𝗻𝗴 𝘁𝗲𝗮𝗺 • Ruflo powers up 𝘁𝗵𝗲 𝗱𝗲𝘃𝗲𝗹𝗼𝗽𝗲𝗿 𝗧𝗵𝗲 𝘄𝗶𝗹𝗱 𝗽𝗮𝗿𝘁: Paperclip only launched in 𝗠𝗮𝗿𝗰𝗵 𝟮𝟬𝟮𝟲. It passed both in about 8 months. 𝗞𝗲𝗲𝗽 𝗶𝗻 𝗺𝗶𝗻𝗱: • 𝗦𝘁𝗮𝗿𝘀 𝗺𝗲𝗮𝘀𝘂𝗿𝗲 𝗮𝘁𝘁𝗲𝗻𝘁𝗶𝗼𝗻, 𝗻𝗼𝘁 𝗿𝗲𝗮𝗹 𝘂𝘀𝗲 • They overlap but aren't the same product. Pick by how you work, not by stars The race isn't to build the best agent anymore. 𝗜𝘁'𝘀 𝘁𝗼 𝗯𝗲 𝘁𝗵𝗲 𝘀𝗰𝗿𝗲𝗲𝗻 𝘆𝗼𝘂 𝗺𝗮𝗻𝗮𝗴𝗲 𝘁𝗵𝗲𝗺 𝗳𝗿𝗼𝗺. 𝘚𝘵𝘢𝘳 𝘤𝘰𝘶𝘯𝘵𝘴 𝘢𝘴 𝘰𝘧 𝘕𝘰𝘷 3, 2026. Links in reply 👇

Just heard decomp.dev is winding down. It's a scoreboard for fan projects that rebuild old games' code. Its maintainer says AI-powered copying broke the community side of the work. But I think the real story is this: 𝗧𝗵𝗲 𝗰𝗼𝗺𝗺𝘂𝗻𝗶𝘁𝘆 𝗲𝘅𝗶𝘀𝘁𝗲𝗱 𝗯𝗲𝗰𝗮𝘂𝘀𝗲 𝘁𝗵𝗲 𝘄𝗼𝗿𝗸 𝘄𝗮𝘀 𝗵𝗮𝗿𝗱. When one game took years, you needed other people. Hard work forced people to work together. AI made the work easy. So people stopped needing each other. 𝘞𝘩𝘢𝘵 𝘰𝘵𝘩𝘦𝘳 𝘤𝘰𝘮𝘮𝘶𝘯𝘪𝘵𝘪𝘦𝘴 𝘰𝘯𝘭𝘺 𝘦𝘹𝘪𝘴𝘵 𝘣𝘦𝘤𝘢𝘶𝘴𝘦 𝘵𝘩𝘦 𝘸𝘰𝘳𝘬 𝘪𝘴 𝘩𝘢𝘳𝘥?

CRAZY: An AI Just Fixed Another AI by Editing it Weights. 𝗟𝗟𝗠𝘀 𝗔𝗿𝗲 𝗦𝘁𝗮𝗿𝘁𝗶𝗻𝗴 𝘁𝗼 𝗥𝗲𝗮𝗱 𝗪𝗲𝗶𝗴𝗵𝘁𝘀 𝗟𝗶𝗸𝗲 𝗖𝗼𝗱𝗲 Here is a question that would have sounded silly not long ago: if you showed an AI the raw numbers inside another AI, could it tell what that AI does? A new post from MIT researchers says: sometimes, yes. Think of a neural network as a giant recipe written in a language nobody speaks. Millions of numbers, no labels. Training writes the recipe. Humans can't really read it. The researchers gave OpenAI's GPT-6 Astra these numbers as text and asked it questions. 𝗪𝗵𝗮𝘁 𝗵𝗮𝗽𝗽𝗲𝗻𝗲𝗱 • 𝗜𝘁 𝗿𝗲𝗮𝗱: shown 4 game-playing networks, it correctly named each game. • 𝗜𝘁 𝗿𝗲𝗽𝗮𝗶𝗿𝗲𝗱: a stuck car-driving network worked after it changed 3.5% of the numbers. Success went from 𝟬 𝘁𝗼 𝟭,𝟬𝟮𝟰 𝗼𝘂𝘁 𝗼𝗳 𝟭,𝟬𝟮𝟰. • 𝗜𝘁 𝗰𝗿𝗲𝗮𝘁𝗲𝗱: with no practice, it wrote a Breakout player that scored 𝟱𝟭𝟳, far above the normally trained networks in the authors' tests (25.3). • 𝗜𝘁 𝗱𝗶𝗮𝗴𝗻𝗼𝘀𝗲𝗱: it found why a small language model kept getting stuck in loops and fixed it. Math scores went from 𝟳.𝟵% 𝘁𝗼 𝟰𝟴.𝟭%. 𝗧𝗵𝗲 𝗼𝘁𝗵𝗲𝗿 𝘀𝗶𝗱𝗲 • It couldn't tell which of two models was trained to flatter users, until it was allowed to see where each model "paid attention". • On a different small model with no looping problem, it couldn't improve math at all. 𝗜𝘁 𝗿𝗲𝗺𝗼𝘃𝗲𝗱 𝗮 𝗿𝗼𝗮𝗱𝗯𝗹𝗼𝗰𝗸. 𝗜𝘁 𝗱𝗶𝗱𝗻'𝘁 𝘁𝗲𝗮𝗰𝗵 𝗮 𝗻𝗲𝘄 𝘀𝗸𝗶𝗹𝗹. • These are small models, in a blog post, tested by the authors themselves. 𝗪𝗵𝗮𝘁 𝘄𝗲 𝗸𝗻𝗼𝘄 A general AI, used as it is, did things with model weights that would be very hard for a person. We don't know yet whether it works on models the size of the ones we use every day. 𝘐𝘧 𝘈𝘐 𝘤𝘢𝘯 𝘳𝘦𝘢𝘥 𝘢𝘯𝘥 𝘦𝘥𝘪𝘵 𝘰𝘵𝘩𝘦𝘳 𝘈𝘐𝘴, 𝘸𝘩𝘰 𝘤𝘩𝘦𝘤𝘬𝘴 𝘵𝘩𝘦 𝘦𝘥𝘪𝘵𝘴? Links in reply 👇

𝗟𝗲𝘁 𝗖𝗹𝗮𝘂𝗱𝗲 𝗧𝘂𝗿𝗻 𝗔𝗻𝘆 𝗛𝗮𝗿𝗱 𝗧𝗼𝗽𝗶𝗰 𝗜𝗻𝘁𝗼 𝗮 𝗖𝗹𝗶𝗰𝗸𝗮𝗯𝗹𝗲 𝗠𝗮𝗽, 𝗪𝗶𝘁𝗵 𝘁𝗵𝗲 𝗣𝗿𝗼𝗺𝗽𝘁𝘀 𝗕𝘂𝗶𝗹𝘁 𝗜𝗻 I have seen a better way in Claude recently. 𝗖𝗵𝗮𝘁 𝗶𝘀 𝗮 𝘀𝘁𝗿𝗮𝗶𝗴𝗵𝘁 𝗹𝗶𝗻𝗲. 𝗬𝗼𝘂 𝗼𝗻𝗹𝘆 𝗹𝗲𝗮𝗿𝗻 𝘄𝗵𝗮𝘁 𝘆𝗼𝘂 𝘁𝗵𝗶𝗻𝗸 𝘁𝗼 𝗮𝘀𝗸, 𝗮𝗻𝗱 𝘄𝗵𝗲𝗻 𝗮 𝘁𝗼𝗽𝗶𝗰 𝗶𝘀 𝗻𝗲𝘄 𝘁𝗼 𝘆𝗼𝘂, 𝘆𝗼𝘂 𝗱𝗼𝗻'𝘁 𝗸𝗻𝗼𝘄 𝘄𝗵𝗮𝘁 𝘁𝗼 𝗮𝘀𝗸. Ask it for a map of a topic, and it draws a flowchart. The topic is split into steps in a logical order: what comes first, what comes next. 𝗔𝗻𝗱 𝗲𝘃𝗲𝗿𝘆 𝗯𝗼𝘅 𝗶𝘀 𝗰𝗹𝗶𝗰𝗸𝗮𝗯𝗹𝗲. Tap one, and Claude writes the next question for you. Usually it sends straight away. Sometimes it lands in your typing box and you just press send. Claude answers that part only, often with a new flowchart. Then you tap again. 𝗟𝗲𝘁'𝘀 𝗧𝗿𝘆 𝗜𝘁: 𝗧𝗿𝗮𝗻𝘀𝗳𝗼𝗿𝗺𝗲𝗿𝘀 Say you want to learn how a transformer works. That's the design from the 2017 paper "𝘈𝘵𝘵𝘦𝘯𝘵𝘪𝘰𝘯 𝘐𝘴 𝘈𝘭𝘭 𝘠𝘰𝘶 𝘕𝘦𝘦𝘥". You ask for a map. You get boxes something like this: • 𝗧𝗼𝗸𝗲𝗻𝘀: your text is cut into small pieces • 𝗘𝗺𝗯𝗲𝗱𝗱𝗶𝗻𝗴𝘀: each piece becomes a list of numbers, plus its position • 𝗔𝘁𝘁𝗲𝗻𝘁𝗶𝗼𝗻: each piece looks at the others • 𝗙𝗲𝗲𝗱-𝗳𝗼𝗿𝘄𝗮𝗿𝗱: each piece is processed on its own • 𝗟𝗮𝘆𝗲𝗿𝘀: attention and feed-forward repeat many times • 𝗡𝗲𝘅𝘁 𝘁𝗼𝗸𝗲𝗻: the model scores every possible next piece Attention sounds like the magic part, so you tap it. You didn't type anything. The question about attention is written for you, and Claude draws a new chart just for that part: • 𝗤𝘂𝗲𝗿𝘆, 𝗸𝗲𝘆, 𝘃𝗮𝗹𝘂𝗲: three versions of each piece • 𝗦𝗰𝗼𝗿𝗲𝘀: compare queries with keys • 𝗦𝗼𝗳𝘁𝗺𝗮𝘅: turn scores into weights • 𝗠𝗶𝘅: a weighted mix of the values • 𝗠𝗮𝗻𝘆 𝗵𝗲𝗮𝗱𝘀: several attentions running side by side Tap "Query, key, value", and it goes one level deeper. 𝗔𝗻𝗱 𝗼𝗻 𝗶𝘁 𝗴𝗼𝗲𝘀. 𝗪𝗵𝘆 𝗜 𝗟𝗶𝗸𝗲 𝗧𝗵𝗶𝘀 • 𝗬𝗼𝘂 𝘀𝗲𝗲 𝘁𝗵𝗲 𝗺𝗮𝗽 𝗯𝗲𝗳𝗼𝗿𝗲 𝘁𝗵𝗲 𝗱𝗲𝘁𝗮𝗶𝗹𝘀. • 𝗘𝗮𝗰𝗵 𝗯𝗼𝘅 𝗶𝘀 𝗮 𝗾𝘂𝗲𝘀𝘁𝗶𝗼𝗻 𝘆𝗼𝘂 𝗱𝗶𝗱𝗻'𝘁 𝗵𝗮𝘃𝗲 𝘁𝗼 𝘄𝗿𝗶𝘁𝗲. • 𝗜𝗳 𝗘𝗻𝗴𝗹𝗶𝘀𝗵 𝗶𝘀𝗻'𝘁 𝘆𝗼𝘂𝗿 𝗳𝗶𝗿𝘀𝘁 𝗹𝗮𝗻𝗴𝘂𝗮𝗴𝗲: it isn't mine. Phrasing the follow-up takes real effort. A tap skips that. 𝗧𝗵𝗲 𝗛𝗼𝗻𝗲𝘀𝘁 𝗣𝗮𝗿𝘁 • 𝗜𝘁'𝘀 𝗻𝗲𝘄: Claude's in-chat visuals launched in beta in March 2026. Other apps may draw a static image or just text. • 𝗧𝗮𝗽𝘀 𝗮𝗿𝗲𝗻'𝘁 𝗰𝗼𝗻𝘀𝗶𝘀𝘁𝗲𝗻𝘁: sometimes the question sends straight away. Sometimes it lands in your typing box and you press send. • 𝗜𝘁 𝘄𝗼𝗿𝗸𝘀 𝗯𝗲𝘀𝘁 𝗶𝗻 𝗖𝗹𝗮𝘂𝗱𝗲'𝘀 𝗰𝗵𝗮𝘁 𝗮𝗽𝗽. 𝗜𝗻 𝗖𝗹𝗮𝘂𝗱𝗲 𝗖𝗼𝗱𝗲, 𝗶𝘁 𝗶𝘀𝗻'𝘁 𝗮𝘀 𝘀𝗺𝗼𝗼𝘁𝗵. • 𝗬𝗼𝘂𝗿 𝗯𝗼𝘅𝗲𝘀 𝘄𝗶𝗹𝗹 𝗱𝗶𝗳𝗳𝗲𝗿: the chart above is an example, not a fixed output. • 𝗧𝗵𝗲 𝗿𝗮𝗯𝗯𝗶𝘁 𝗵𝗼𝗹𝗲: four charts later, you can lose where you started. Ask for the path at the top of every chart. • 𝗧𝗼𝗼 𝗻𝗲𝗮𝘁: a flowchart can make a messy topic look simpler than it is. 𝗪𝗵𝗲𝗿𝗲 𝗶𝘁 𝘄𝗼𝗿𝗸𝘀: 𝗖𝗹𝗮𝘂𝗱𝗲 𝗰𝗵𝗮𝘁. 𝗪𝗵𝗲𝗿𝗲 𝗶𝘁 𝗱𝗼𝗲𝘀𝗻'𝘁: 𝗖𝗹𝗮𝘂𝗱𝗲 𝗖𝗼𝗱𝗲. 𝗧𝗵𝗲𝗿𝗲, 𝗶𝘁 𝗱𝗼𝗲𝘀𝗻'𝘁 𝘄𝗼𝗿𝗸 𝗮𝘀 𝗲𝘅𝗽𝗲𝗰𝘁𝗲𝗱 𝗮𝘁 𝗮𝗹𝗹. 𝗧𝗿𝘆 𝗜𝘁 𝗬𝗼𝘂𝗿𝘀𝗲𝗹𝗳 (𝗦𝘄𝗮𝗽 𝘁𝗵𝗲 𝗕𝗿𝗮𝗰𝗸𝗲𝘁𝘀) Give me a clickable map of [topic] as a flowchart. Use 5 to 7 boxes, in the order I should learn them. Make every box send its follow-up message as soon as I tap it, so I never type anything or press send. That message should ask about that step only and draw a new flowchart for it the same way. Show the path at the top of every chart (like: Transformer > Attention > Query, key, value) so I don't get lost. 𝘚𝘵𝘰𝘱 𝘨𝘶𝘦𝘴𝘴𝘪𝘯𝘨 𝘵𝘩𝘦 𝘯𝘦𝘹𝘵 𝘲𝘶𝘦𝘴𝘵𝘪𝘰𝘯. 𝘈𝘴𝘬 𝘧𝘰𝘳 𝘵𝘩𝘦 𝘮𝘢𝘱. Links in reply 👇

hackertube.org/data The #7 talk: youtube.com/watch?v=qCGofLIz… John Sotos, DEF CON 25: youtube.com/watch?v=HKQDSgBH… Chuck Moore: youtube.com/watch?v=0PclgBd6…

𝗞𝗮𝗿𝗽𝗮𝘁𝗵𝘆'𝘀 𝗛𝗮𝗰𝗸𝗲𝗿𝗧𝘂𝗯𝗲 𝘃𝘀 𝗬𝗼𝘂𝗧𝘂𝗯𝗲: 𝗥𝗲𝗺𝗲𝗺𝗯𝗲𝗿𝗲𝗱 𝘃𝘀 𝗖𝗹𝗶𝗰𝗸𝗲𝗱. There are two very different ways to decide if a video is worth your time. 𝗬𝗼𝘂𝗧𝘂𝗯𝗲 𝗰𝗼𝘂𝗻𝘁𝘀 𝗰𝗹𝗶𝗰𝗸𝘀. Views, watch time, what keeps you scrolling. 𝗛𝗮𝗰𝗸𝗲𝗿 𝗡𝗲𝘄𝘀 𝗰𝗼𝘂𝗻𝘁𝘀 𝘀𝗼𝗺𝗲𝘁𝗵𝗶𝗻𝗴 𝘀𝗹𝗼𝘄𝗲𝗿: someone coming back, months or years later, to tell a stranger "𝘺𝘰𝘶 𝘩𝘢𝘷𝘦 𝘵𝘰 𝘸𝘢𝘵𝘤𝘩 𝘵𝘩𝘪𝘴." Most of the time, the two agree. I wanted to find where they don't. 𝗪𝗵𝗮𝘁 𝗜 𝗱𝗶𝗱 Karpathy just launched HackerTube. Claude read 19 years of Hacker News comments and kept the YouTube videos HN keeps recommending: 𝟲,𝟰𝟱𝟰 𝗼𝗳 𝘁𝗵𝗲𝗺, ranked by how many different months people recommended each one. The data is public. It has HN's side of the story, not YouTube's. So I got a free YouTube API key and pulled the view count for all 6,454. 𝗪𝗵𝘆? If HN's favourites are also YouTube's most-watched, great videos find their audience. If not, some of the best talks are getting lost. 𝗖𝗼𝗺𝗽𝗮𝗿𝗶𝗻𝗴 𝗹𝗶𝗸𝗲 𝘄𝗶𝘁𝗵 𝗹𝗶𝗸𝗲 HackerTube sorts every video into 𝟭𝟰 𝘁𝘆𝗽𝗲𝘀: talks, lectures, explainers, demos, comedy, documentaries, music videos and more. Those types live in different worlds. The typical 𝗺𝘂𝘀𝗶𝗰 𝘃𝗶𝗱𝗲𝗼 in the collection has 𝗮𝗯𝗼𝘂𝘁 𝟲.𝟭 𝗺𝗶𝗹𝗹𝗶𝗼𝗻 𝘃𝗶𝗲𝘄𝘀. The typical 𝗲𝘅𝗽𝗹𝗮𝗶𝗻𝗲𝗿 has about 2.7 million. The typical 𝘁𝗮𝗹𝗸 has about 85,000. So comparing a 50-minute conference talk with Despacito would prove nothing. I kept only 𝘁𝗮𝗹𝗸𝘀 𝗮𝗻𝗱 𝗹𝗲𝗰𝘁𝘂𝗿𝗲𝘀 (𝟭,𝟳𝟲𝟱 𝘃𝗶𝗱𝗲𝗼𝘀), then only the ones 𝟱 𝗼𝗿 𝗺𝗼𝗿𝗲 different HN commenters recommended. That left 391. The typical one has 𝗮𝗯𝗼𝘂𝘁 𝟮𝟳𝟮,𝟬𝟬𝟬 𝘃𝗶𝗲𝘄𝘀. So usually, good talks do get found. 𝗕𝘂𝘁 𝗻𝗼𝘁 𝗮𝗹𝘄𝗮𝘆𝘀 𝗧𝗵𝗲 #𝟳 𝘃𝗶𝗱𝗲𝗼 𝗼𝘂𝘁 𝗼𝗳 𝗮𝗹𝗹 𝟲,𝟰𝟱𝟰 is Armin Ronacher's "𝘏𝘰𝘸 𝘗𝘺𝘵𝘩𝘰𝘯 𝘸𝘢𝘴 𝘚𝘩𝘢𝘱𝘦𝘥 𝘣𝘺 𝘭𝘦𝘢𝘬𝘺 𝘐𝘯𝘵𝘦𝘳𝘯𝘢𝘭𝘴." 𝟭𝟱 𝗱𝗶𝗳𝗳𝗲𝗿𝗲𝗻𝘁 𝗛𝗡 𝗰𝗼𝗺𝗺𝗲𝗻𝘁𝗲𝗿𝘀 recommended it. It has 𝗮𝗯𝗼𝘂𝘁 𝟮𝟬,𝟬𝟬𝟬 𝘃𝗶𝗲𝘄𝘀. That's more than 13 times below a typical talk at the same level. It's not alone: • 𝗝𝗼𝗵𝗻 𝗦𝗼𝘁𝗼𝘀 𝗮𝘁 𝗗𝗘𝗙 𝗖𝗢𝗡 𝟮𝟱: 11 recommenders, 𝗮𝗯𝗼𝘂𝘁 𝟭𝟯,𝟵𝟬𝟬 𝘃𝗶𝗲𝘄𝘀 • 𝗖𝗵𝘂𝗰𝗸 𝗠𝗼𝗼𝗿𝗲 𝗼𝗻 𝗮 𝟭𝟰𝟰-𝗰𝗼𝗺𝗽𝘂𝘁𝗲𝗿 𝗰𝗵𝗶𝗽: 7 recommenders, 𝗮𝗯𝗼𝘂𝘁 𝟭𝟵,𝟭𝟬𝟬 𝘃𝗶𝗲𝘄𝘀 • 𝗠𝗼𝘅𝗶𝗲 𝗠𝗮𝗿𝗹𝗶𝗻𝘀𝗽𝗶𝗸𝗲 𝗮𝘁 𝗗𝗘𝗙 𝗖𝗢𝗡 𝟭𝟴: 6 recommenders, 𝗮𝗯𝗼𝘂𝘁 𝟮𝟭,𝟳𝟬𝟬 𝘃𝗶𝗲𝘄𝘀 𝗙𝗼𝗿 𝘃𝗶𝗲𝘄𝗲𝗿𝘀: the view count tells you what's popular, not what people still remember years later. These four are worth a weekend. 𝗙𝗼𝗿 𝗯𝘂𝗶𝗹𝗱𝗲𝗿𝘀: HackerTube's data is free to reuse. Joining it with one more source took one API key and about 130 requests. 𝗧𝗼 𝗯𝗲 𝗳𝗮𝗶𝗿 • 𝗧𝗵𝗲 𝗱𝗮𝘁𝗮 𝘀𝗵𝗼𝘄𝘀 𝘁𝗵𝗲 𝘃𝗶𝗲𝘄𝘀 𝗮𝗿𝗲 𝗹𝗼𝘄, 𝗻𝗼𝘁 𝘄𝗵𝘆. Long technical talks on small channels may simply have a small audience. • 𝗛𝗮𝗰𝗸𝗲𝗿𝗧𝘂𝗯𝗲'𝘀 𝗵𝗼𝗺𝗲𝗽𝗮𝗴𝗲 𝗮𝗹𝗿𝗲𝗮𝗱𝘆 𝗺𝗶𝘅𝗲𝘀 𝗶𝗻 "𝗵𝗶𝗱𝗱𝗲𝗻 𝗴𝗲𝗺𝘀." What's new here is the numbers. • I searched YouTube for other copies of each talk. I dropped one (Guy Steele's "Growing a Language") because 𝗮𝗻𝗼𝘁𝗵𝗲𝗿 𝘂𝗽𝗹𝗼𝗮𝗱 𝗵𝗮𝘀 𝗮𝗯𝗼𝘂𝘁 𝟭𝟲𝟱,𝟬𝟬𝟬 𝘃𝗶𝗲𝘄𝘀. • The types and "recommenders" are labels written by Claude from HN comments. Counts are as of 2 Nov 2026. 𝗩𝗶𝗲𝘄𝘀 𝘁𝗲𝗹𝗹 𝘆𝗼𝘂 𝘄𝗵𝗮𝘁'𝘀 𝗽𝗼𝗽𝘂𝗹𝗮𝗿. 𝗛𝗮𝗰𝗸𝗲𝗿 𝗡𝗲𝘄𝘀 𝘁𝗲𝗹𝗹𝘀 𝘆𝗼𝘂 𝘄𝗵𝗮𝘁 𝗽𝗲𝗼𝗽𝗹𝗲 𝘀𝘁𝗶𝗹𝗹 𝗿𝗲𝗺𝗲𝗺𝗯𝗲𝗿. 𝘞𝘩𝘪𝘤𝘩 𝘩𝘪𝘥𝘥𝘦𝘯 𝘵𝘢𝘭𝘬 𝘸𝘰𝘶𝘭𝘥 𝘺𝘰𝘶 𝘢𝘥𝘥 𝘵𝘰 𝘵𝘩𝘪𝘴 𝘭𝘪𝘴𝘵? Links in reply 👇

𝗠𝘆 𝗬𝗼𝘂𝗧𝘂𝗯𝗲 𝗛𝗼𝗺𝗲𝗽𝗮𝗴𝗲 𝘃𝘀 𝗛𝗮𝗰𝗸𝗲𝗿𝗧𝘂𝗯𝗲 I put my YouTube homepage next to Karpathy's HackerTube. Top 6 videos on each. 𝗧𝗲𝘅𝘁 𝗮𝗱𝗱𝗲𝗱 𝗼𝗻 𝘁𝗵𝗲 𝘁𝗵𝘂𝗺𝗯𝗻𝗮𝗶𝗹: • YouTube: 𝟲 𝗼𝗳 𝟲 • HackerTube: 𝟬 𝗼𝗳 𝟲 "?" 𝗼𝗿 "!" 𝗼𝗻 𝘁𝗵𝗲 𝘁𝗵𝘂𝗺𝗯𝗻𝗮𝗶𝗹: • YouTube: 𝟯 𝗼𝗳 𝟲 • HackerTube: 𝟬 𝗼𝗳 𝟲 Part of that is by design. HackerTube doesn't show the creator's thumbnail at all. It shows 𝗮 𝗳𝗿𝗮𝗺𝗲 𝗳𝗿𝗼𝗺 𝘁𝗵𝗲 𝗺𝗶𝗱𝗱𝗹𝗲 𝗼𝗳 𝘁𝗵𝗲 𝘃𝗶𝗱𝗲𝗼. Creators design the thumbnail to get the click. 𝗡𝗼𝗯𝗼𝗱𝘆 𝗱𝗲𝘀𝗶𝗴𝗻𝘀 𝘁𝗵𝗲 𝗵𝗮𝗹𝗳𝘄𝗮𝘆 𝗳𝗿𝗮𝗺𝗲. And these are mostly channels I chose to follow. Even the people I like are playing the thumbnail game, because that's what the algorithm rewards. To be fair: • One of my 6 is an ad. • HackerTube still shows YouTube's own titles. One of its 6 is "𝘛𝘩𝘦 𝘴𝘱𝘦𝘦𝘤𝘩 𝘵𝘩𝘢𝘵 𝘣𝘳𝘰𝘬𝘦 𝘵𝘩𝘦 𝘪𝘯𝘵𝘦𝘳𝘯𝘦𝘵 - 𝘔𝘰𝘴𝘵 𝘐𝘯𝘴𝘱𝘪𝘳𝘪𝘯𝘨 𝘦𝘷𝘦𝘳." • My homepage is personal. Yours will look different. 𝘞𝘩𝘢𝘵 𝘥𝘰𝘦𝘴 𝘺𝘰𝘶𝘳 𝘠𝘰𝘶𝘛𝘶𝘣𝘦 𝘩𝘰𝘮𝘦𝘱𝘢𝘨𝘦 𝘭𝘰𝘰𝘬 𝘭𝘪𝘬𝘦 𝘳𝘪𝘨𝘩𝘵 𝘯𝘰𝘸? 𝘗𝘰𝘴𝘵 𝘢 𝘴𝘤𝘳𝘦𝘦𝘯𝘴𝘩𝘰𝘵 👇

𝗔𝗙𝗧𝗘𝗥 𝗧𝗛𝗘 𝗖𝗟𝗢𝗡𝗘 𝗙𝗟𝗢𝗢𝗗, 𝗔𝗡𝗧𝗛𝗥𝗢𝗣𝗜𝗖 𝗦𝗧𝗘𝗣𝗦 𝗜𝗡 A few weeks ago REA (Reverse Engineer Anything) and similar Claude skills went viral. Soon my feed was full of reverse engineered clones of famous games and software, each one shared with a GitHub link. Now Anthropic has stepped in. Around 21 October 2026 people saw Claude refusing reverse engineering work. Around 27 October 2026 a new rule went into Claude Code telling it not to recreate someone else's game, even with a new name. Addy Osmani from Anthropic later confirmed it. You can still build your own games. I get the copyright side. But some developers say Claude now refuses their own projects halfway through, and that is a real problem. Anthropic's confirmation, via Game Developer: gamedeveloper.com/programmin…