https://open.video/@velum-theory/v/the-ai-model-that-refuses-to-write-a-word The AI Model That Refuses to Write a Word TypeSafe launched Jev, a System One model that returns typed decisions instead of writing text. In this video, I explain why the key benchmark is not just accuracy but the shape of the question: 62.6% for one question versus 95.0% when the same model is asked five narrow questions and scored with a regression. The result is a model that works faster, cheaper, and with a constrained output schema. This is a technical explainer about the Jev architecture, the calibration idea, and the tradeoff between decision-only models and general text generation. https://video-meta.open.video/poster/URzy3xDQth-j/ytpYlu1Qa62_bnrotC.jpg https://streaming.open.video/contents/URzy3xDQth-j/1790448940/index.m3u8 https://open.video/embed?contentId=ytpYlu1Qa62 298 https://open.video/@velum-theory/v/how-quantum-computers-actually-work-and-why-theyre-faster How Quantum Computers Actually Work (and Why Theyre Faster) Quantum computers are not faster laptops. They are faster at a few specific jobs, and this video builds one of those jobs from scratch, with numbers you can check by hand. We start with a puzzle: four boxes, one hidden prize, lids shut. An ordinary computer has one move -- open a box and look -- so it averages two and a half opens for four boxes, and half a million for a million boxes. Then we build the machine that can do better. A bit is a switch. A qubit keeps a number for each answer, called an amplitude, and squaring it gives the chance (0.71 squared is a half). The pair of numbers is an arrow whose length is always one, so the chances always add to 100%. An amplitude can also be NEGATIVE, and a measurement cannot tell plus from minus -- a fact that decides everything later. Two qubits keep four amplitudes: one per box. But measuring collapses all of it to one random box, so the numbers have to be rearranged first, using interference. The Hadamard gate is shown twice over: from zero it gives +0.71 and +0.71 from one it gives +0.71 and MINUS 0.71. Run it twice and four paths appear the two paths to zero give 0.5 + 0.5 = 1, while the two paths to one give 0.5 - 0.5 = 0. The wrong answer erases itself. Then Grovers search on the four boxes: start at a half each, let one check flip the prizes sign, flip every amplitude around the average of 0.25, and box three goes to 1 while the empties fall to 0 -- the right answer, after a single check. Finally: inside a real machine (the dilution refrigerator, its temperature stages, the wiring and the chip), a million boxes in about 785 rounds, Shors algorithm and encryption, and the honest limits. CHAPTERS 00:00 Five minutes vs ten septillion years 00:30 The puzzle: four boxes, one prize 01:22 What a qubit really is 02:45 Two qubits, four numbers, four boxes 03:18 The catch: measuring throws it away 03:54 Interference and the Hadamard gate 05:16 Solving the puzzle in one check 06:29 Inside a real quantum computer 07:42 A million boxes, and breaking encryption 08:21 Why it https://video-meta.open.video/poster/Tuv46DKsBxea/yJpdR0uQb72_xwgmFa.jpg https://streaming.open.video/contents/Tuv46DKsBxea/1790449546/index.m3u8 https://open.video/embed?contentId=yJpdR0uQb72 559 https://open.video/@velum-theory/v/how-transformers-actually-work-the-math-with-real-numbers How Transformers Actually Work: The Math, With Real Numbers A modern language system runs on one equation: softmax of Q times K transpose over the square root of d, times V. This video builds the whole pipeline from the bottom and works every step by hand, with small numbers you can check with a pen. Byte pair encoding (merging e+s, es+t, lo+w on a real toy corpus) one-hot times the embedding matrix as a lookup words as points and king - man + woman the dot product (3x2 + 4x5 = 26) and cosine similarity (0.97) sinusoidal positional encoding query, key and value scores 6, 5, 1.5, 0.4 scaled by the square root of 2 softmax to 64% / 32% / 3% / 1% the weighted sum that moves bank to (1.69, 0.37) QK transpose and the causal mask multi-head attention, residual connections, layer norm and the feed-forward network logits, softmax and temperature on a five-word vocabulary cross-entropy loss, the p - y gradient, and gradient descent. CHAPTERS 00:00 One equation behind every language model 00:42 Tokens: byte pair encoding by hand 01:41 One-hot times the embedding matrix 02:14 Words as points: king - man + woman 02:46 The dot product and cosine similarity 03:42 Positional encoding: sines and cosines 04:31 One word, two meanings 04:57 Query, key, value 05:29 Scores and the square root of d 06:12 Softmax, worked out 06:52 The weighted sum: bank moves 07:24 All words at once: QK transpose and the mask 08:10 The transformer block: heads, residual, norm, FFN 09:26 Logits, softmax and temperature 10:20 Loss, gradient, gradient descent 11:21 The whole chain SOURCES & NOTES Vaswani et al., Attention Is All You Need, 2017 (arXiv 1706.03762): scaled dot-product attention, sinusoidal positional encoding, multi-head attention, residual connections with layer norm, and the position-wise feed-forward layer. Sennrich, Haddow & Birch, Neural Machine Translation of Rare Words with Subword Units, 2016 (arXiv 1508.07909): byte pair encoding the low/lower/newest/widest corpus is their worked example. Mikolov et al., 2013 (arXiv 1301.3781): word2vec and the king - man + woman analogy. Radford et https://video-meta.open.video/poster/lazK6nbl-tTG/ytVtl117bR2_jxVCCL.jpg https://streaming.open.video/contents/lazK6nbl-tTG/1790449565/index.m3u8 https://open.video/embed?contentId=ytVtl117bR2 718 https://open.video/@velum-theory/v/langchain-explained-how-ai-apps-are-really-built LangChain Explained: How AI Apps Are Really Built What is LangChain, and why do so many AI apps use it? A language model on its own is brilliant with words, but it has never read your documents, cannot use your tools, and forgets every conversation. In this video we follow a support chatbot for an online shop and see how LangChain snaps prompts, models, parsers, retrievers and tools together like building blocks, and how agents, LangGraph and LangSmith take it all the way to production. In this video: Why a language model alone is like a brilliant new hire on day one The glue code problem: spaghetti code and rewrites for every model provider LangChains big idea: every part of an AI app is a block with the same shape Your first chain: prompt template model output parser, and invoke() Swapping models (ChatGPT, Claude, a local model) by changing one line RAG, the open-book exam: loaders, splitters, embeddings, vector stores, retrievers Agents: a chain is a train on rails, an agent is a taxi driver LangGraph: flowcharts with save points, human approval and memory LangSmith: the flight recorder for tracing and testing AI apps Who uses it (Klarna, Uber, Elastic), and when NOT to use LangChain CHAPTERS 00:00 Why a smart model is not enough 00:37 A language model: a brilliant new hire 01:03 The glue code problem 01:28 LangChain: building blocks for AI apps 02:01 Your first chain: prompt, model, parser 02:42 Swap any model with one line 03:12 RAG: the open-book exam 03:41 Embeddings, vector stores and retrievers 04:18 Agents: rails vs routes 05:06 LangGraph: save points, approvals, memory 05:48 LangSmith: the flight recorder 06:19 From side project to unicorn 06:53 When NOT to use LangChain 07:24 The verdict SOURCES LangChain Blog (22 Oct 2025): LangChain and LangGraph Agent Frameworks Reach v1.0 Milestones (createagent no breaking changes until 2.0). LangChain Blog (20 Oct 2025): LangChain raises 125M to build the platform for agent engineering. TechCrunch (21 Oct 2025): Open source agentic startup LangChain hits 1.25B valuation. Wikipedia: LangChain (launched October 2 https://video-meta.open.video/poster/kgDC7nqsBdMi/4dUsRe0kGR2_VjHKZA.jpg https://streaming.open.video/contents/kgDC7nqsBdMi/1790449552/index.m3u8 https://open.video/embed?contentId=4dUsRe0kGR2 482 https://open.video/@velum-theory/v/cicd-explained-how-real-teams-ship-code-without-breaking-things CI/CD Explained: How Real Teams Ship Code Without Breaking Things Every team that ships software eventually hits the same wall: a change that worked on one laptop takes the live site down. This video explains CI/CD as one story, from a Friday-evening deploy gone wrong to a bug fix that goes from push to live in twenty minutes, with every step checked by a machine. In this video: How teams used to ship: merge day, integration hell, midnight checklists Continuous integration: merge small, and let a pipeline check every change One push followed start to finish, from a red cross to a green merge The real GitHub Actions file behind it, line by line Continuous delivery, and the manual deploy that cost Knight Capital 440 million Continuous delivery vs continuous deployment Safety nets: rollbacks, gradual rollouts and feature flags Where CI/CD came from: Extreme Programming, CruiseControl, Jenkins, GitHub Actions How pipelines go wrong: slow pipelines, flaky tests, thin tests, leaked secrets CHAPTERS 00:00 The Friday deploy 00:48 How teams used to ship 01:25 Continuous integration 02:05 One push, start to finish 02:51 The file behind it 03:27 Continuous delivery (and a 440M mistake) 04:08 Delivery vs deployment 04:46 The whole pipeline 05:23 Shipping safely 06:04 Where CI/CD came from 06:56 Where it goes wrong 07:36 Your first pipeline SOURCES Martin Fowler, Continuous Integration (martinfowler.com) Devopedia, Continuous Integration: the phrase first appears in Grady Boochs 1991 book Kent Beck made it a core practice of Extreme Programming in the late 1990s. Kohsuke Kawaguchi, Bye bye Hudson, hello Jenkins (kohsuke.org, Jan 2011) Jenkins (software), Wikipedia: first Jenkins release February 2011. Jez Humble and David Farley, Continuous Delivery (Addison-Wesley, 2010) continuousdelivery.com, Continuous Delivery vs Continuous Deployment (Aug 2010). GitHub Blog, GitHub Actions now supports CI/CD (Aug 8, 2019) TechCrunch (Aug 8, 2019): general availability November 13, 2019. U.S. SEC order In the Matter of Knight Capital Americas LLC (Oct 2013) and case studies: a manual deployment missed https://video-meta.open.video/poster/Scj4VnftFtfO/4dUtRfvBHQ2_KBAPmG.jpg https://streaming.open.video/contents/Scj4VnftFtfO/1790449545/index.m3u8 https://open.video/embed?contentId=4dUtRfvBHQ2 496 https://open.video/@velum-theory/v/docker-explained-how-containers-really-work-zero-to-pro Docker Explained: How Containers Really Work (Zero to Pro) What is Docker, and how does it actually work? No Docker or DevOps knowledge needed. We follow one small pizza ordering app that works on my machine but crashes on the server, and fix it step by step, from the very first container all the way to Kubernetes. In this video: Why apps break when they move between computers The old fixes: long setup guides and heavy virtual machines The shipping container idea behind Docker Containers vs virtual machines (apartments vs houses) Images vs containers, explained with a song file The Dockerfile recipe, layers and build caching Running a container and opening a port Sharing images through a registry like Docker Hub Volumes: keeping your data when containers come and go Docker Compose: running a whole app with one command Under the hood: the Docker engine, namespaces, control groups, layered file systems Kubernetes: running thousands of containers Pro tips: small images, no baked-in secrets, no root, scanning CHAPTERS 00:00 It works on my machine 00:41 Why apps break when they move 01:09 The old fixes: setup guides and virtual machines 01:37 The shipping container idea 02:17 Containers vs virtual machines 02:51 Images and containers 03:21 The Dockerfile: a recipe for an image 03:55 Layers and caching 04:25 Running a container and opening a port 05:02 Sharing images: registries and Docker Hub 05:37 Volumes: keeping your data 06:10 Docker Compose: many containers together 06:48 How Docker really works: namespaces and control groups 07:38 Kubernetes: thousands of containers 08:21 Pro tips: small, safe images 08:57 The whole journey SOURCES Wikipedia: Docker (software) Docker blog, 11 Years of Docker (first demoed by Solomon Hykes at PyCon, 15 March 2013). 2025 Stack Overflow Developer Survey, Technology section (Docker used by 71% of respondents, +17 points). Port Houston: the Ideal X (26 April 1956, 58 containers). Docker docs: Dockerfile reference, multi-stage builds, volumes, networking, Docker Compose, Docker Desktop WSL 2 backend. Linux man pages and kernel docs: namespace https://video-meta.open.video/poster/RznuNzu6wtLO/y6VsQ0uRbR2_sfOtqO.jpg https://streaming.open.video/contents/RznuNzu6wtLO/1790450920/index.m3u8 https://open.video/embed?contentId=y6VsQ0uRbR2 576 https://open.video/@velum-theory/v/the-internet-is-torturing-a-fruit-fly-it-isnt The Internet Is Torturing a Fruit Fly (it isnt) A complete map of a fruit flys nervous system went public, and within a week the internet had it playing Doom, parallel parking, and trading bitcoin off its own dopamine neurons. This video breaks down what the dataset actually contains, what those browser demos are really running, and whether any of it can suffer. Everything on screen is a real capture: the papers, the repos, the live demos, the reporting. Sources are listed below. CHAPTERS 00:00 The map that took a decade 00:24 What dropped: the male connectome 01:00 What were checking 01:13 What a connectome actually is 01:45 The model bolted on top 02:10 How much fly is really running 02:39 Is anything being tortured? 03:10 What is genuinely good here 03:25 Go poke at it SOURCES MaleCNS v1.0 (adult male central nervous system: central brain, optic lobes, ventral nerve cord), HHMI Janelia FlyEM with Google Research and collaborators more than 166,000 neurons and roughly 125 million synapses paper published 3 September 2026. janelia.org/project-team/flyem FlyWire / FAFB adult female brain connectome: 139,255 proofread neurons and about 50 million connections, the first complete wiring diagram of an adult brain, published in Nature, October 2024 proofreading was crowdsourced. flywire.ai Becky Ferreira, A Digital Fly Brain Has Taken Over the Internet, 404 Media, 15 September 2026: the Minecraft, Doom, Beat Saber, parallel parking and bitcoin projects, and the researchers position that a connectome is not sentient. awesome-fly (CC0), a curated list of fly connectome projects, including the 80-neuron dino game, the 6,000-neuron swat game and the browser simulation of all 139,255 neurons. github.com/cobanov/awesome-fly Fruit fly photograph: Sanjay Acharya, Wikimedia Commons, CC BY-SA 4.0. Note: facts are as of September 2026. Not affiliated with HHMI Janelia, Google, 404 Media or any project shown. Screenshots are used for commentary and identification. Fruit fly photograph by Sanjay Acharya (CC BY-SA 4.0). Icons: Tabler Icons (MIT) logos: Simple Icons (CC0). connec https://video-meta.open.video/poster/inr8JnGYYtLG/yApsQK16a62_dqpYNS.jpg https://streaming.open.video/contents/inr8JnGYYtLG/1790450902/index.m3u8 https://open.video/embed?contentId=yApsQK16a62 223 https://open.video/@velum-theory/v/mcp-vs-api-explained-how-ai-agents-use-your-apps MCP vs API Explained: How AI Agents Use Your Apps Everyone in AI is talking about MCP. Some call it the new API others say it will make APIs obsolete. Both are wrong. In this video we start from what an API really is, show the tangle of custom connectors that AI apps used to need, and explain how the Model Context Protocol works: one standard USB-C port that lets any AI assistant discover and use tools, usually by calling ordinary APIs underneath. In this video: What an API is: a service window with its own menu and rules The tangle: why 5 AI apps and 5 tools meant 25 custom connectors MCP, the USB-C port for AI: 25 connectors become 10 plugs How it works: host, client and server tools, resources and prompts The big difference: tools the AI discovers at run time, described for the AI to read One request, start to finish: booking a meeting and emailing the agenda MCP vs API side by side, and why MCP is an adapter, not a replacement Adoption: OpenAI, Google, Microsoft, 10,000+ servers, and the Linux Foundation The risks: tool poisoning, prompt injection, and how to stay safe CHAPTERS 00:00 Why AI assistants had no hands 00:32 What an API is 01:09 The tangle: 25 custom connectors 01:39 MCP: one plug for AI 02:13 How MCP works: host, client, server 02:45 The big difference: tools the AI can discover 03:18 One request, start to finish 03:52 MCP vs API, side by side 04:24 Everyone plugged in (2024 to 2026) 04:58 Risks: tool poisoning and prompt injection 05:20 The verdict SOURCES Anthropic (Nov 2024): Introducing the Model Context Protocol. modelcontextprotocol.io: What is the Model Context Protocol? (Think of MCP like a USB-C port for AI applications). MCP specification, revision 2026-07-28, and the MCP architecture docs (host, client, server tools, resources, prompts). TechCrunch (Mar 2025): OpenAI adopts Anthropics standard for connecting AI models to data. MCP Blog and the Linux Foundation (9 Dec 2025): MCP joins the Agentic AI Foundation 10,000+ active public MCP servers. OWASP: MCP Tool Poisoning. Cloud Security Alliance research note (Jul 2026): MCP tool poison https://video-meta.open.video/poster/UojSUpXQsFpJ/yAVdQ0u7a62_vHbVgX.jpg https://streaming.open.video/contents/UojSUpXQsFpJ/1790450909/index.m3u8 https://open.video/embed?contentId=yAVdQ0u7a62 355 https://open.video/@velum-theory/v/unix-vs-linux-whats-the-real-difference Unix vs Linux: Whats the Real Difference? Open a terminal on a Mac and another on a Linux server, type the same commands, and you get almost the same results. Yet the Mac is officially certified as Unix, and Linux, technically, is not. So what is the real difference? In this video we follow the whole story, from Bell Labs in 1969 to the phone in your pocket, and explain why Linux is Unix-like, why your Mac really is Unix, and why the two still feel like twins. In this video: What an operating system does: the kernel is the engine, the shell and tools are the rest of the car 1969: Ken Thompson and Dennis Ritchie create Unix at Bell Labs, then rewrite it in C The Unix way: small tools that each do one thing well, joined by pipes BSD, commercial Unix, and the Unix wars GNU: Richard Stallmans free rebuild (GNUs Not Unix) 1991: Linus Torvalds writes a kernel just a hobby and Linux is born Is Linux a Unix? Unix-like vs certified UNIX (and why macOS passes) POSIX: the shared rulebook that makes your skills carry over Who runs the world today: supercomputers, the cloud, Android, Mars CHAPTERS 00:00 Twins: a Mac and a Linux server 00:28 What an operating system does 00:57 1969: Unix is born at Bell Labs 01:32 The Unix way: small tools and pipes 02:00 BSD, big business and the Unix wars 02:35 GNU: the free rebuild begins 03:09 1991: Linus Torvalds and Linux 03:43 Is Linux a Unix? (Unix-like vs certified) 04:23 POSIX: one shared rulebook 04:52 Who runs the world today 05:29 The whole picture SOURCES The Open Group, Register of UNIX Certified Products: macOS 26 Tahoe, UNIX 03 (certificate P1223, first issued 29 Aug 2025). Huawei (Sep 2016): KunLun EulerOS 2.0 passes UNIX 03 certification. Linus Torvalds, comp.os.minix, 25 Aug 1991: What would you like to see most in minix? Richard Stallman, the GNU announcement (27 Sep 1983), gnu.org. D. M. Ritchie and K. Thompson, The UNIX Time-Sharing System, Communications of the ACM, 1974. D. M. Ritchie, The Evolution of the Unix Time-sharing System, 1979. IEEE Std 1003.1 (POSIX) the name POSIX was suggested by Richard Stallman. https://video-meta.open.video/poster/k9rO6nWJAd-i/56oIluuka62_jQZpJU.jpg https://streaming.open.video/contents/k9rO6nWJAd-i/1790451424/index.m3u8 https://open.video/embed?contentId=56oIluuka62 365 https://open.video/@velum-theory/v/cursors-origin-the-end-of-github Cursors Origin: The End of GitHub? On August 17, 2026, GitHub suffered an outage of more than six hours, and on that very same day, Cursor launched Origin, its own home for code. So is this the end of GitHub? In this video we tell two origin stories: how four MIT students rebuilt Microsofts own editor around AI and turned it into one of the fastest-growing developer tools ever, and why Cursor is now building a code host made for AI agents. In this video: Writing code vs storing code: the editor (workshop) and the code host (library) How GitHub became the home of open source, and why Microsoft paid 7.5 billion for it Cursors origin: four MIT students, a fork of VS Code, and a bet that AI belongs at the heart of the editor Tab, codebase chat and agents: why developers switched From startup to a 60 billion SpaceX deal The new bottleneck: code review, and the Graphite acquisition Origin: repositories, pull requests, cloud agents and two-way GitHub sync The verdict: is this really the end of GitHub? CHAPTERS 00:00 The day GitHub stumbled 00:27 Two jobs: writing code vs storing code 01:00 GitHubs empire (and Copilot) 01:33 Cursors origin: four MIT students 02:03 Why developers switched 02:29 From startup to 60 billion 02:59 The new bottleneck: code review 03:27 Origin: Cursors home for code 04:04 The verdict: the end of GitHub? SOURCES TechCrunch (18 Aug 2026): Cursor capitalizes on GitHub frustration, launches rival hosting platform. SiliconANGLE (17 Aug 2026): Cursor launches Origin code hosting service to compete with GitHub. DigitalOcean: Cursor Origin vs. GitHub, the 2026 code-hosting face-off. TechCrunch / CNBC (16 Jun 2026): SpaceX to acquire Cursor for 60B in stock. Forbes (8 Jun 2026): Cursor hits 4 billion annualized revenue. TechCrunch (19 Dec 2025): Cursor acquires code review startup Graphite. TechCrunch (11 Oct 2023): Anysphere raises 8M from OpenAI to build an AI-powered IDE. GitHub Octoverse 2025: 180M+ developers on GitHub. GeekWire (Aug 2026): GitHub outage disrupts developers worldwide. Note: facts are as of September 2026. Origin is https://video-meta.open.video/poster/oPjKRn17EdDb/z6Uckv0BaQ2_VYUbSl.jpg https://streaming.open.video/contents/oPjKRn17EdDb/1790451437/index.m3u8 https://open.video/embed?contentId=z6Uckv0BaQ2 283 https://open.video/@velum-theory/v/every-api-explained-rest-graphql-grpc-websockets-webhooks Every API Explained: REST, GraphQL, gRPC, WebSockets & Webhooks Every time you check the weather, pay online or log in with Google, dozens of programs quietly talk to each other through APIs. But there isnt just one kind of API. In this video we meet every major type, and see exactly how each one talks, using everyday analogies: a service window with a menu, a certified letter, a custom order form, a shared codebook, an open phone call, a radio station and a doorbell. In this video: What an API really is: a contract, and a window in a wall APIs without the internet: library functions and system calls How web APIs talk: HTTP requests and responses, status codes (200, 404, 500) and JSON REST: resources, GET / POST / PUT / DELETE, and why stateless scales SOAP: XML envelopes and strict contracts (still big in banks and airlines) GraphQL: one address, and exactly the fields you ask for gRPC: Protocol Buffers, compact binary and streaming over HTTP/2 Real time: polling vs WebSockets vs Server-Sent Events (how AI chat streams its answers) Webhooks: when the server calls you API keys, OAuth tokens and rate limits (429 Too Many Requests) The API map: choose any API with two simple questions CHAPTERS 00:00 Programs talking to programs 00:26 What is an API? The service window 01:03 APIs without the internet: libraries & system calls 01:37 How web APIs talk: HTTP, status codes, JSON 02:23 REST: the counter with a menu 03:06 SOAP: the certified letter 03:32 GraphQL: the custom order form 04:09 gRPC: the shared codebook 04:45 Real time: polling, WebSockets, Server-Sent Events 05:29 Webhooks: dont call us, well call you 05:59 API keys, OAuth tokens and rate limits 06:28 The API map: every style in one picture 07:06 Every API in one breath SOURCES & FURTHER READING Roy T. Fielding, Architectural Styles and the Design of Network-based Software Architectures, doctoral dissertation, UC Irvine, 2000 (chapter 5: REST). W3C: SOAP Version 1.2 and WSDL 1.1. GraphQL Foundation: graphql.org (specification, and the history of GraphQL at Facebook). grpc.io documentation, and protobuf.dev (Proto https://video-meta.open.video/poster/O9jCUpeIBF-k/zQotRvuAbQ2_DpJwNE.jpg https://streaming.open.video/contents/O9jCUpeIBF-k/1790451391/index.m3u8 https://open.video/embed?contentId=zQotRvuAbQ2 459 https://open.video/@velum-theory/v/linux-file-system-explained-everything-is-a-file Linux File System Explained: Everything Is a File Why does Linux have no C: drive? Because everything, from your documents to your disks, your keyboard and even running programs, lives in one single tree that starts at /. In this video we explore the Linux file system like a city, one district at a time. In this video: Why Linux uses one tree instead of drive letters Everything is a file what it really means A guided tour of /home, /root, /etc, /usr, /boot, /var, /tmp, /dev, /proc, /mnt, /lib and /opt Absolute vs relative paths, and the . .. shortcuts What a file really is: inodes, hard links and symbolic links Permissions made simple: rwx, owner/group/others and 755 Mounting a USB stick, file systems, and how the ext4 journal survives a power cut CHAPTERS 00:00 No C: drive? Welcome to Linux 00:26 One single tree, starting at / 01:01 Everything is a file 01:33 City tour: /home, /root, /etc, /usr, /boot 02:30 City tour: /var, /tmp, /dev, /proc, /mnt, /lib, /opt 03:21 Paths: absolute, relative, . .. and 03:56 Inodes, hard links and symbolic links 04:45 Permissions: rwx and 755 05:28 Mounting, file systems and the ext4 journal 06:06 Recap: the whole map SOURCES & FURTHER READING Filesystem Hierarchy Standard, version 3.0 (The Linux Foundation). Linux man pages: hier(7), pathresolution(7), inode(7), proc(5), chmod(1), mount(8). Linux kernel documentation: ext4 data structures and the journal (kernel.org). freedesktop.org: the /usr merge (why /bin points into /usr/bin). Note: details vary a little between distributions. For example, most (not all) current distros cleanly merge /bin into /usr/bin, clear /tmp at boot, and use ext4 by default (Fedora, for one, uses Btrfs). Watch next: What Is Bio Computing? https://youtu.be/8jFJ5K1Ob9Y Linux LinuxForBeginners FileSystem https://video-meta.open.video/poster/OPzi-nvYEB-O/5Aodl0v7a72_vRpDBa.jpg https://streaming.open.video/contents/OPzi-nvYEB-O/1790451435/index.m3u8 https://open.video/embed?contentId=5Aodl0v7a72 403 https://open.video/@velum-theory/v/this-computer-is-alive-what-is-bio-computing This Computer Is Alive: What Is Bio Computing? 800,000 living brain cells, grown in a dish, learned to play Pong in about 5 minutes and nobody programmed them. Welcome to bio computing: computers built from the materials of life. In this video: Why your 20-watt brain outclasses a 20-megawatt supercomputer on efficiency How Leonard Adleman solved a maths puzzle in a test tube in 1994 How scientists program bacteria to behave like logic gates Brain organoids on electrode chips, and the first biological computer you can buy The hard problems: lifespan, speed, consistency and ethics CHAPTERS 00:00 Brain cells that learned to play Pong 00:19 What is a computer, really? 00:55 The 20-watt brain vs a 20-megawatt supercomputer 01:25 The three branches of bio computing 01:40 DNA computing: a puzzle solved in a test tube 02:43 Living circuits: programming bacteria 03:30 Living neurons: brain organoids on a chip 04:36 Why its not in your phone (yet) 05:17 The future: silicon + biology 05:42 Recap: what is bio computing? SOURCES & FURTHER READING Kagan et al. (2022). In vitro neurons learn and exhibit sentience when embodied in a simulated game-world. Neuron. (DishBrain / Pong) Cortical Labs: CL1 biological computer announcement (March 2025). Jordan et al. (2024). Open and remotely accessible Neuroplatform for research in wetware computing. Frontiers in Artificial Intelligence. (FinalSpark) Adleman, L. M. (1994). Molecular computation of solutions to combinatorial problems. Science. Erlich & Zielinski (2017). DNA Fountain enables a robust and efficient storage architecture. Science. Gardner, Cantor & Collins (2000). Construction of a genetic toggle switch in Escherichia coli. Nature. Elowitz & Leibler (2000). A synthetic oscillatory network of transcriptional regulators. Nature. (the repressilator) Smirnova et al. (2023). Organoid intelligence (OI): the new frontier in biocomputing and intelligence-in-a-dish. Frontiers in Science. TOP500 list: power draw of the Frontier supercomputer (21 MW). Note: 200 million gigabytes per gram is a theoretical maximum fo https://video-meta.open.video/poster/VPvGIBHJEhwj/zQoZkL0Qb62_XgVAga.jpg https://streaming.open.video/contents/VPvGIBHJEhwj/1790451383/index.m3u8 https://open.video/embed?contentId=zQoZkL0Qb62 367