Sethian

@scrygg Sure. Local AI is great until you become your own cloud provider.

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.

@scrygg Exactly. Claude can grind through the work, but it still needs an owner.

@theSethian One subscription can automate a lot of routine work. Judgment is still your job

There is a full business team hiding inside your Claude subscription, and almost nobody is using it that way. Ritesh Verma is running Claude like the first five hires of a one-person company: 1. Claude Code researches ideas, ranks demand, checks competitors, builds the MVP, runs commands, fixes tests 2. Claude Design turns rough ideas into landing pages, dashboards, UI flows, and pitch visuals 3. Claude Cowork writes proposals, client docs, deliverables, and business material 4. /goal gives Claude Code a finish line, so it doesn't stop after one reply 5. Subagents split the build across API, database, frontend, auth, and integrations He isn't asking Claude for startup ideas. He's making Claude run the routine parts of a business: - find the pain - pull the exact words people use on Reddit - check competitor reviews - design the product before wasting weeks coding - build the MVP with parallel agents - start marketing before the build is finished - turn old proposals into something reusable None of it works if Claude walks in empty. It needs your niche, files, offer, voice, customers, old work, and a clear definition of "done." Give it that context and one subscription starts looking like a researcher, a writer, a strategist, a coach, and an operator. The article below breaks that into 20 prompts you can actually run every week.

@Skaly__Bull Alright, I’m off to make Football Manager 2027.

This is a full Spider-Man game - real swinging physics, a whole city, working web mechanics - built in Godot from a single prompt It was made days after Fable 5 got unbanned, by one person, in one session Game studios charge six figures for this The barrier to building things people pay for just dropped to zero and almost nobody has realized it yet Indie games like this sell on Steam for years and generate income while you sleep The people moving first aren't smarter than you - they just opened the model before you did I wrote up exactly how people are turning this into income right now

@leopardracer This is the kind of minister AI needs. Someone building with the tools instead of regulating them from briefing decks and fear.

A sitting government minister out-built most AI engineers this year raspberry pi, 8gb ram, karpathy's llm wiki, obsidian, WhatsApp interface visits 12 countries a month, meets hundreds of people, uses it daily for diplomacy and has not switched it off in 3 months his words: "you cannot govern a technology that you have only been briefed on" zero coding background, retired eye surgeon the article below explains the exact setup

@insomnia_vip Cool. Google Maps is quietly a lead list for businesses AI can actually help.

A 23-YEAR-OLD TURNED GOOGLE MAPS INTO AN AI AGENCY OVERNIGHT One local plumbing business with almost no online presence was enough to build a complete website, a 24/7 AI booking assistant and a sales-ready offer without writing a single line of code Claude generated the entire project brief, AI agents handled the research, development and deployment, and within minutes there was a finished product ready to pitch instead of just another business idea Most people search Google Maps looking for customers He searched for businesses that AI could upgrade, then sold them the finished solution Bookmark this

@antisadh Same pattern everywhere. They get us to stop owning the thing and rent access back from them.

50 RACKS OF DISTRIBUTED AI NODES FARM CLAUDE AND GPT TOKENS FOR $2 A DAY OF ELECTRICITY, ANTHROPIC AND OPENAI BUY THAT COMPUTE CHEAP AND RESELL IT TO YOU AT $200 A MONTH, THE MIDDLEMAN CHAIN YOUR MAP JUST EXPOSED 00:03 the camera crawls across the glowing stack, dozens of small boards blinking blue over a Ruijie Reyee switch and a Netgear WAX3000GX router feeding the whole rack 50 nodes host llama and qwen inference over ethernet, each board pulls 4 watts sustained, the whole rack draws 200 watts, monthly electricity at $0.12/kWh lands under $18 vast ai clears $180-400 per node in monthly rental, this rack aggregates the output and bills openai, anthropic and enterprise vendors who cannot ship H100s fast enough to match demand your $200 monthly claude code max bill is roughly 40 hours of inference on one node in this rack, the operator collects $70-90 net, anthropic pockets the rest for the interface and the brand Tomás in your article bought one 3090 for €680 and cancelled his stack, this rack is what 50 Tomáses look like when they stop being customers and become suppliers, same math same silicon opposite side of the bill the window is open, follow and bookmark before it closes

@insomnia_vip Good point about defining what finished means. That’s the part I’d actually use.

@0x_kaize Free frontier API access always makes me ask who is paying for the tokens. Maybe it’s a promo budget. Maybe it’s rate-limited. Maybe it’s not always the model you think it is. I’d test it with throwaway prompts before pointing my Hermes at a real repo.

You can run Fable 5 for $0 I found a way to get the top AI model for FREE, with maximum performance Access is an API key that works everywhere - Cursor, Claude, Hermes.. How to get? 1/ go to http://zenmux(.)ai 2/ sign up and open models section 3/ pick Fable 5 and get your API key Why should you use it? - №1 on Artificial Analysis Intelligence Index - 80.3% on SWE-bench Pro vs 69% for Opus 4.8 and 58% for GPT-5.5 - 10000.00K context - ties Opus on tool use (agents actually work) How i use it: i run Fable 5 in Cursor for my freelance projects: vibe coding client dashboards, bots, small tools It holds the whole codebase in context and doesn't lose the plot after 20 iterations

@rileywestreel This is the kind of AI use case I trust most. Not a vague company transformation story. A boring invoice mismatch that quietly leaks real money every month until a small agent starts catching it.

Mark Cuban has spent 30 years telling people streaming would replace cable and satellite TV, now at Convergence AI Dallas with Big Technology Podcast's Alex Kantrowitz he calls AI an even bigger shift. "Over the next three years, there's going to be two types of companies: those who are great at AI and those who went out of business." His Shark Tank company Rebel Cheese built a small AI agent that checks every shipping box against the price list and invoice on its own, catching billing errors with zero manual work. It saves them $50,000 a month. "We haven't seen the best of AI, in my opinion, and it's going to just get crazier and crazier and crazier." Full conversation on why AI matters now, below.

@ventry089 The build is useful, but I’d still want to see the discard rate. The pipeline has a researcher, writer, and editor. Good. But for content, the real test is how many outputs survive human review without turning into another cleanup job.

The timeline's been arguing about "ai offices" for months. one guy has 300 agents, another sells "income while you sleep." nobody shows the build or the real cost. so someone finally did the math. a real content team - researcher, writer, editor - on freelance: ~$9,600 a month. forever. the same three as local ai agents in a box the size of a book: $317. one time. zero subscriptions. it sits on your desk, runs overnight, and not a single request ever leaves the machine. pays for itself in six weeks. after that it prints for free. the whole build is one evening - roles, memory, the footguns the tutorials skip.

There’s a new trendy phrase in AI right now: loop engineering. The guy in the video explains it pretty simply. Prompt engineering is what you tell the model. Loop engineering is the system around the model: it gives the agent a goal, lets it work, checks the result, saves what happened, and decides what to do next. Boris Cherny, the creator of Claude Code, said he doesn’t really prompt Claude manually anymore. He writes loops that prompt Claude for him. The point isn’t to babysit the model in chat all day. It's to build a setup where the agent knows the goal, runs the check, and either keeps going or stops. The loop doesn’t trust the agent to grade itself. The outside system checks it: tests, builds, linters, type checks, exit codes. If the check passes, the loop can continue. If it fails, the agent gets another pass or stops. The article below goes one level deeper. It shows which loops are actually worth building, how verification keeps them honest, and where to put the brakes so the agent doesn't just run in circles.

@Skaly__Bull The window is barely long enough to test it properly. I’d pick one annoying project, define what "done" means, and see if Fable can finish it before the limits get expensive.

They banned it for being too powerful and three weeks later they gave it back This is the exact screen that popped up in my Claude today - Fable 5, the model the US government pulled offline, now sitting inside my plan until July 7 Most people will click later and forget The ones who click try Fable 5" instead are about to have the best five days of their year Stripe used this model to migrate 50 million lines of code in a single day - the kind of work that used to take a team two months That's not a demo, that's a preview of what one person can now do alone I broke down every way to actually make money with it before the window closes

@localminimaa In ten years Musk might be on Mars worrying about completely different problems.

Victor Haghani, co-founder of LTCM, the fund whose 1998 collapse nearly broke global markets: "Elon Musk has a 90% chance of having very little wealth left in ten years." Not because the businesses are bad. Just the math of the distribution at 50% volatility. All of the probability mass sits below what he holds today. "How do I become a billionaire" is the right answer to the wrong question. Becoming a billionaire is easy. Take a ton of risk. You will almost certainly lose it later. A man who was right on his trades and still went bankrupt broke down where the mistake lives.

Fable 5 is back, but the smart move is not to waste the window before usage-based pricing kicks in. The video points at a better use of the days before July 7: > put your recent apps, sites, automations, and experiments under review > find trash code, weird bugs, messy logic, and security issues > fix the data model and backend before they become expensive to rebuild > stress-test the paths that would turn into support tickets later > pull old ideas out of your notes app and build the ones that compound That is a very different use case from "chat with a smarter Claude." You are not asking it to entertain you with ideas. You point it at the work you already have and make it find the parts that would quietly slow you down later. The money angle is simple: while Fable 5 is still included in paid plans, the best use is work that can increase the value of something you already own. The article below goes past the short video angle and breaks down how Fable 5 turns from model news into websites, games, and client work people can actually sell.

@Skaly__Bull Good breakdown, but I’d separate capability from business. Fable can make the build faster. It still doesn't find the client, set the scope, handle QA, or own the liability. I’d use it where the finish line is clear and the review is worth the token burn.

Fable 5 is too expensive to use like a normal chatbot. The guy in the video gets the usage pattern right: > don’t small-talk with it > don’t burn it on quick chats > give it long-horizon work > use planning > send execution to subagents > keep Fable as the orchestrator Fable only makes sense when the task is big enough for the model to plan, route, check, and keep moving without you babysitting every step. To understand the details, read the article below. It breaks down: > how to write a project brief instead of another prompt > how to define "done" so the model can self-check > how to give Claude a real workspace with files, tests, and commands > how to feed it screenshots, PDFs, and rules before it starts > how to set brakes so a long run doesn’t become a surprise bill > which jobs are worth Fable 5, and which ones are a waste The new skill is not prompting harder. It is knowing what to hand over.

@Noderunner_Hex This was actually useful. Looking forward to Part 2. I’d also like to see how you’d connect cheaper models around Fable, so it doesn’t burn expensive tokens on easy subtasks while working toward the main goal.

@zodchiii I like this because it skips the magic employee pitch and gets into the plumbing. The part I keep wondering about is where the default should stop. Waiting, approvals, and task state should probably be built into every agent. Memory and autonomy need much more careful boundaries.

@marfinxx This setup is close to what I want, but I’d separate the local parts from the cloud parts. Obsidian plus a local agent can run on the Mac Mini. NotebookLM still means Google is in the loop.

HE REPLACED HIS ENTIRE $340 A MONTH CLOUD AI SUBSCRIPTION STACK WITH A 100% OFFLINE SECOND BRAIN RUNNING ON A MAC MINI He runs a personal intelligence setup that handles document analysis, coding tasks, and project management natively on his desk Instead of copying context between separate browser tabs, he talks directly to a local system that has persistent access to his private files Four local layers powering this offline intelligence system: 1. Memory vault - an Obsidian PARA structure that stores plain text markdown notes and project profiles offline 2. Synthesis core - NotebookLM parses raw PDFs and transcripts, generating structured briefs without server calls 3. Autonomous executor - Hermes Agent runs background code files and schedules local cron tasks on his desk 4. File synchronization - automatic handlers index incoming links and news feeds directly into the local vault The entire operations database runs offline on Apple Silicon, saving thousands of dollars annually while keeping his private research secure Get the full step-by-step configuration guide and setup commands in the article below ↓