Corvus

@yulaxbt That's probably the best second life an old gaming PC can get

YOUR OLD GAMING PC IS NOT DEAD. IT IS A LOCAL AI SERVER He turned a dead gaming PC into a private AI box instead of throwing it away. The Mac is just the clean front end, while the old tower runs headless Linux, local models and the ugly terminal work in the background. The best part is the moment the model starts thinking and the PC actually spins up. That is when it stops feeling like a normal desktop and starts feeling like private AI infrastructure under the desk. This is why local AI keeps getting more interesting. You do not always need a new $2K AI PC. Sometimes the cheapest local setup is the gaming machine you already forgot about.

@antisadh Three months of debugging just to end up with a Mac Mini is kind of wild

3 MONTHS OF SEARCHING FOR ONE MOTHERBOARD WITH TWO PCIE X8 SLOTS AT 75 WATTS EACH FAILED ACROSS 4 CHASSIS, THE $599 MAC MINI M4 IN THE MAP SOLVES THE WHOLE STORAGE PROBLEM WITH A 5GB/S SOLDERED SSD OUT OF THE BOX 02:33 the builder stares down at his Mac mini on the shelf, "this computer down here that has been replaced by this Mac mini, definitely has a faster processor and RAM" he wanted two 6.8TB PCIe SSD expansion cards running full speed for a fast NAS to feed his AI workstation, the search chewed through a returned AMD Epyc build, a Z840 dual Xeon, an HP Omen and a dusty Dell workstation the Epyc had the slots but the wrong wattage, the Z840 had the slots but not the rebar, the HP Omen only had one x16, the Dell had spider webs and three x2 slots, none of the 4 chassis on his shelf ran both cards at full x8 the $599 Mac Mini M4 ships with a 2TB soldered NVMe at 5 GB/s reads, no expansion card required, no PCIe slot math, plug it in and the same NAS use case works over 10GbE thunderbolt in a box the size of a paperback part 4 of the map covers the mac mini as the easy default and it lands there for a reason, the DIY compute-first path burns weeks on chassis debugging that the appliance tier skips entirely the window is open, follow and bookmark before it closes

@Di_Krass_ Interesting point, the feedback loop is probably the bigger issue than AI itself

Adam Aleksic, linguist (TED): "if you start talking more like ChatGPT, you probably start thinking more like ChatGPT. if you don't question yourself, their version of reality becomes your version of reality." at 04:43 he lays the loop bare: AI distorts reality → we repeat it → AI gets fed our distorted data → repeat. we're already inside it. 8 minutes on the word "delve," on hyperpop on Spotify, on Musk hand-tuning Grok, and on why machine language is leaking back out of your mouth. Watch it now, and you'll know more than 99% of people. Follow me.

@pilsnoir Builders move during windows

THE US GOVERNMENT BANNED FABLE 5 FOR 19 DAYS AND THE MOMENT IT CAME BACK ANTHROPIC SET A 6-DAY CLOCK BEFORE $100 BURNS IN 9 MINUTES anthropic dropped Fable 5 on June 9th. Three days later it was blocked globally with no warning before the ban, Stripe used it to migrate 50 million lines of code in one day instead of two months until July 7th it runs inside your current plan. After that it's $10 per million tokens in and $50 out The people building on it right now aren't waiting to find out if it gets banned again

@scrygg Private compute gets messy

LOCAL AI CLUSTERS ARE COOL UNTIL YOU HAVE TO MAINTAIN THEM This video shows the part of local AI that usually gets skipped. A stack of small Mac mini-style machines, terminals, code, browser tests, monitoring dashboards and a developer trying to turn a desk setup into a private compute cluster. The upside is obvious. You own the hardware, keep the workflow local, run experiments on your own machines and avoid depending on cloud GPUs for every task. But the caption nails the tradeoff: upfront cost, maintenance, debugging and networking make this approach much better for serious developers than casual users. That is why the clip works. It shows local AI as it really is right now: powerful, private and exciting, but still messy enough that the people who win first will be the ones willing to own the whole stack.

@lagerskoy Small box, real workstation

THIS MINI PC IS STARTING TO LOOK LIKE A FULL WORKSTATION GMKtec EVO-X2 Ryzen AI Max+ 395 looks like a small desk box, but the video shows it moving through real creator workloads: productivity, DaVinci Resolve, Blender, 2D art, 3D art and workstation-style tasks without making the setup feel heavy. The AI part makes it more interesting. The built-in GMK assistant lets you run local models, translate text, pick models and keep parts of the workflow offline instead of sending every small task to ChatGPT or another cloud tool. That is why this category keeps getting stronger. Mini PCs are no longer just quiet office boxes, they are turning into compact creator machines with enough memory, local AI and desktop power to replace a much bigger setup.

@lagerskoy @claudeai Slack became the interface

CLAUDE IS ALREADY BECOMING A TEAMMATE INSIDE SLACK This video shows the part of AI agents that feels more real than another standalone chatbot. @claudeai lives inside Slack channels, understands the thread, reads the request, checks the context and turns a normal teammate tag into actual work. The example is simple, but that is why it lands. Someone asks Claude to add the blog draft and beta invite email, and instead of opening another app, writing a prompt and pasting context manually, the task happens inside the same workspace where the team already talks. That is the real interface shift. AI agents will not only live in separate chat windows, they will move into Slack, docs, code, inboxes and internal tools, where the context already exists and the work is already happening.

NVIDIA JUST MADE LOCAL AI LOOK CHEAP NVIDIA is showing a tiny AI computer that costs $249, runs at 25W and still pushes almost 70 trillion operations per second. In the video, the whole pitch is simple: this is no longer a huge workstation story, it is a small board you can hold in one hand. The interesting part is what they say it can run. Large language models, AI workloads and the same kind of software stack used on much bigger NVIDIA systems, but inside a device that looks closer to a mini dev board than a serious AI machine. This is why local AI is moving so fast. The hardware is getting smaller, cheaper and more accessible, while the use cases are moving from cloud demos into devices people can actually buy, plug in and build with.

ANTHROPIC IS NOT BUILDING A CHATBOT COMPANY Most people still look at Anthropic through Claude, but this video frames the company as something much bigger: research talent, AI infrastructure, enterprise adoption, security, government-level trust and a valuation moving toward $180B. The edit shows the full story in a very business-heavy way. OpenAI roots, research culture, Stanford-style technical depth, Anthropic’s office, charts going vertical, AI security visuals and the idea that safe AI is becoming one of the most valuable layers in the entire market. That is the part people miss. Claude is the product everyone touches, but the actual bet is control over the AI workflow stack: models, safety, enterprise trust and the infrastructure layer companies will build on top of.

HE DIDN’T BUY A MAC STUDIO. HE BOUGHT A 24/7 AI EMPLOYEE At first this looks like a normal unboxing. A guy opens a Mac Studio box in his kitchen and says his AI agent machine just arrived. But the real point is not the Apple hardware. Once a machine sits on your desk and runs agents all day, it stops being just a computer. It becomes the box that can read files, run scripts, process notes, answer messages, check tasks and keep working after your laptop closes. Most people still rent AI inside a browser tab. The next step is giving that AI a machine to live on. The Mac Studio is the hands. The agent is the worker.

AMD WANTS TO END THE $200 AI SUBSCRIPTION ERA Most people rent AI from the cloud, AMD is showing a tiny desktop box that can run serious models locally with 128GB of shared memory across CPU, GPU and NPU The bigger story is not just the hardware. It is the shift from paying monthly for intelligence to owning the machine that runs it If local AI keeps getting smaller and cheaper, the next advantage will not be who has the best subscription, it will be who has private compute sitting on their desk

MOST PEOPLE ARE BURNING CLAUDE LIMITS FOR NO REASON Everyone complains that Claude Code runs out too fast But the real problem is how they use it. They let one model plan, search, edit, debug and carry the entire chat history until the context gets bloated This video shows the smarter workflow: use Opus to think through the plan, switch to Sonnet for simple coding, compact old context before it becomes dead weight and make Claude reason before burning tokens on bad edits The best Claude users are not just prompting better They are managing compute better

DARIO AMODEI THINKS OPEN SOURCE AI HAS A CHINA PROBLEM Most people talk about open-source AI like it is only about freedom, speed and innovation Dario is pointing at the darker side. Strong models will keep leaking into the world, including to countries and groups that may use them for cyber operations, surveillance and military advantage The uncomfortable part is that AI does not stay contained once it is open. If a model is powerful enough to accelerate product building, coding and research, it is also powerful enough to accelerate the people you do not want using it That is the real tension Open AI helps everyone move faster, including the wrong people

AMD JUST PUT LOCAL AI INSIDE A BOX SMALLER THAN A CONSOLE Most people think serious AI needs cloud servers, giant GPU rigs or expensive workstations AMD is showing the opposite: Ryzen AI Halo, a compact local AI development system with unified memory shared between CPU, GPU and NPU, built to run models locally without needing to connect to anything The real shift is not the device itself, it is what it means. AI is moving from remote APIs into personal hardware, and once small machines can run powerful models at your desk, the question changes from “which cloud tool do you rent” to “what can you build when the intelligence is sitting right next to you”

AI AGENTS WERE THE WARMUP, AI LOOPS ARE THE REAL GAME Most people are still talking about agents like they are magic workers, but this video shows the next step: an AI loop that wakes up on a schedule, checks your niche, decides what matters and sends you what to do next The difference is simple. An agent waits for instructions, a loop keeps running without you babysitting it. Every Sunday at 6PM it can scan your world, find the signal, prepare your week and message you before you even open your laptop That is where AI gets interesting, not as a chatbot you visit, but as a system that quietly watches, thinks and pushes useful work back to you

CLAUDE IS TURNING ROBLOX DEV INTO A ONE PERSON GAME STUDIO Most people think building a Roblox game means learning scripting, UI, game loops, rewards, inventories and endless bug fixing This video shows a different workflow. One creator is using Claude to build and iterate a full Roblox game, from the world and UI to rolling systems, rewards and backend logic, while he just keeps testing, giving feedback and pushing the game closer to launch The crazy part is not that AI can write code, it is that game development is starting to look like direction instead of implementation. The winners may not be the best coders anymore, but the people who can spot addictive mechanics and use AI to ship them fast

YOUR NEXT AI COMPUTER MIGHT LOOK LIKE A WIFI ROUTER Most people still imagine local AI as a giant workstation with loud fans, huge GPUs and cables everywhere, but this video shows the opposite: a tiny GIGABYTE AI TOP ATOM box sitting on a desk and being positioned as a personal AI supercomputer The interesting part is not just the size, it is the shift in what computers are becoming. PCs used to be built for browsing, gaming and editing, now a new category is forming around running models, agents and AI workflows locally without depending on cloud servers Cloud AI made intelligence feel like a subscription, machines like this make it feel like hardware again, your desk, your compute, your models, your data

HERMES AGENT JUST MADE BUILDING AI AGENTS FREE FOREVER Most AI agent tools still lock the best parts behind subscriptions, credits or API bills, this video shows Hermes going the opposite direction Free agent API, local engine, Claude Code-style building, agent workflows and a system where you can start creating AI agents without paying just to experiment The important shift is not that one tool became free, it is that the cost of building agents keeps collapsing. When the tools become free, the advantage moves from access to execution Soon the winners will not be the people who can afford AI agents They will be the people who actually know what to build with them

CLAUDE IS TURNING RANDOM PRODUCT IDEAS INTO PREMIUM WEBSITES Most people still think AI websites look like cheap templates. Then you watch Claude turn a simple product concept into a full brand experience with 3D visuals, bold typography, recipes, product sections and animations that feel like they came from a real studio. The crazy part is not just the design. It is how fast the gap between idea and execution is collapsing. A founder can now sit down with Claude in the morning and have something that looks investor-ready before most agencies even send the first moodboard. This is where web design is heading. Less waiting, fewer meetings, faster launches and a lot more people building brands before they ever hire a team.

@lagerskoy Excellent presentation of the material

A KID JUST MADE 15 MILLION ROBUX IN 2 MONTHS WHICH IS $52,500 USD IN REAL CASH FROM A GAME HE BUILT IN 3 HOURS USING FREE MODELS WITH ZERO LINES OF CODE. THE IDEA IS THE ONLY THING THAT MATTERED AND HE FOUND IT WHILE SCROLLING TIKTOK IN BED. Here's exactly what happened. He was lying in bed scrolling TikTok. Saw a video of someone using aimbot in a shooter game. Thought to himself "I'd love to try cheating in video games but my account would get banned." Then the idea hit him. What if he built a Roblox game where the cheats were the entire game? Where every player gets aimbot. Where the "exploit experience" is the official feature. Where you can hit headshots without consequences because it's the literal point of the game. He opened Roblox Studio. Searched the free models marketplace. Pulled an FPS template. Pulled the Crossroads classic map. Stitched them together. Added auto-aim mechanics. Shipped in 3 hours. The game looked like absolute garbage. He admits it on camera. The graphics were rough. The mechanics were buggy. The UI was generic. It didn't matter. Aimbot Arena hit 5,000 concurrent players within weeks. The game cleared $52,500 in DevEx earnings in 2 months. From a 3 hour build. With zero engineering work. Just free models and a strong idea. Here's the math nobody is doing. 3 hours of work yielded $52,500. That's $17,500 per hour of effective work. A senior Google engineer makes $200,000 a year. This kid out-earned 3 months of Google senior salary in a single afternoon. Now do the deeper math. Niche selection is 70 percent of the outcome on Roblox. The other 30 percent is thumbnail and shipping speed. The mechanic of "let players cheat legally" was a niche worth finding. The kid found it scrolling TikTok in bed. He shipped it the same day. Every gamer over 15 has wondered what cheating in a game would feel like. The market wanted this. Nobody had built it. He showed up first. This is the part most people will miss. The barrier to building successful Roblox games in 2026 is not technical skill. It is taste. The ability to scroll cultural feeds, spot the gap between "what people want to experience" and "what currently exists", and ship something that closes that gap before anyone else notices. A 15 year old who studies TikTok, YouTube, and Roblox trends for 30 minutes a day is more likely to find a 15 million Robux idea than a senior Unity engineer who builds technically perfect games for years. The economics of attention beat the economics of effort every single time on this platform. If you have a kid between 12 and 18, this is the platform to point them at. If you are between 18 and 30 and you have spent any time watching Roblox content, you already have everything you need. Open Roblox Studio. Pull free models. Ship something this weekend. The next viral concept is sitting in your TikTok feed right now waiting for someone to notice it.