raiden

@AnatoliKopadze he is genius, thanks for share

Anthropic engineer: "Fable 5 is already smarter than we know how to use. The bottleneck was never the AI, it's you." In 19 minutes he shows exactly how to get everything out of Claude with no extra tools, no extra costs. You're already paying for all of this, just not using it. Watch the session, then read the guide below on the Claude features 99% of users never find.

THIS CHINESE BUILDER TURNED $2,800 OF USED RTX 3090s INTO A 96GB AI RIG THAT KILLED HIS $459 A MONTH SUBSCRIPTIONS AND RENTS ITSELF OUT FOR $1,580 A MONTH WHILE HE SLEEPS. Kai is 29 and writes backend code in Shenzhen. He had never added up his AI bill until five renewals hit in one week. Claude Code Max, ChatGPT Pro, Gemini, Copilot, Cursor. $459 a month to rent compute on the AI companies' servers that a single $500 graphics card can replace. Then he understood. He could run all of it himself and save hundreds a month. Four used RTX 3090s off Xianyu at about $700 each, 24GB apiece, bolted into an open frame on his bedroom floor. Three commands and Ollama was running the same models the subscriptions rented him. He cancelled everything except one ChatGPT Plus at $20. The $459 a month stopped. Then he took it further. A couple of the cards cover his own work, so he lists the rest on a GPU rental marketplace and lets strangers run their AI on them overnight. The same companies that sell you a $200 subscription rent their compute from rooms exactly like his. He stopped paying them and started charging. He built it once for $2,800. Now it clears about $1,580 a month and he barely touches it. He used to pay to run AI on someone else's servers. Now other people pay to run theirs on his. Save this before the slot closes.

@yurshevv this is gem bro

Most people are wasting money on AI without realizing it. Here's a simple example. Imagine you have a 30,000-token system prompt. Now imagine your app calls Claude 20 times during a single workflow. Most developers pay to process those same 30,000 tokens... 20 different times. That's 600,000 input tokens before any real work even begins. Prompt Caching changes the equation. Instead of paying for the same context over and over, Claude processes it once and reuses it across future requests. Lower cost. Lower latency. Exactly the same behavior. This is why AI is moving beyond prompt engineering. The biggest gains no longer come from writing better prompts. They come from designing better systems. How you structure context. What stays constant. What changes. Where you split memory from instructions. The companies building the best AI products aren't obsessing over prompts anymore. They're obsessing over architecture. Prompt engineering was the first wave. Context engineering is the next one.

@eng_khairallah1 bro this share in insane

A 17-year-old makes $10,400 a week. From Google Maps. She's never built a website. I laughed. She showed me her Stripe. All she does is open Google Maps. Types "tacos Philadelphia." Finds one with 4.7 stars and a site from 2013. Drops their info into a site builder. It spits out the whole thing. 20 minutes. 6 months ago she was making $9/hr at McDonald's. Still has the uniform in her closet. She just sends the finished site. Goes to sleep. Their name already on it. That taco shop owner opened the link. Showed his daughter. "Mira, this is us?" Paid $4,800 that night. "hey is this real? how much" — that was her Tuesday. A web agency: $14,500. Four people. Six weeks. Her: $0. Alone. 20 minutes. First month: $800. Second month: $4,100. Last week: $10,400. Her dad still thinks she works at McDonald's. She found it on Google Maps. So can you. This is the same complete system to build a $10K/month B2B lead machine using the same logic - fully automated, no cold calling, no ads ↓ Bookmark this. This is the one.

@sairahul1 this actually makes sense, prompt engineering was always temporary anyway

🚨 CEO of Nvidia: "Nobody writes prompts anymore. The new job is to write and handle loops." This is the shift that's going to define the rest of 2026. 53 minutes of pure insight from one of the richest men on earth. Watch it, then read the full guide on how to actually create loops below

@cyrilXBT bro if this is real thats actually insane

this is f*cking gold Andrej Karpathy came over to Anthropic just five weeks back. Someone on his team pulled up the actual Claude.md they run day to day and showed it to me. I plugged it straight into my workflow. Claude’s next reply wasn’t just improved. It felt like a different model. The canned, one-size-fits-all stuff vanished, and it started reasoning in my voice. Save this before it disappears down your timeline. Read it first then hit the article below.

@yurshevv this is actually insane

While everyone is chasing better AI models... Andrej Karpathy is building something far more valuable. A second brain that gets smarter every day. Not another chatbot. Not another note-taking app. A system that remembers. Every PDF. Every YouTube video. Every research paper. Every business idea. Every conversation. Here's why I think this changes everything. Most people use AI like this: Open ChatGPT. Ask a question. Get an answer. Close the tab. Tomorrow? Start over. You never build leverage. Karpathy's workflow is the opposite. Every new source is added to a living knowledge base. AI connects ideas. Updates old conclusions. Finds contradictions. Links everything together. Instead of answering one question... ...it improves the entire system. Think about what that means. After 12 months, your AI doesn't just know more. It knows you better. Your business. Your writing. Your research. Your mistakes. Your thinking. I don't think the biggest advantage in AI will come from using GPT-6 or Claude 5. It will come from owning years of structured context that nobody else has. Models will become cheaper. Models will become faster. Models will become available to everyone. Your accumulated knowledge won't. The people who start building that today will have an advantage that's almost impossible to copy in a few years. Most people are collecting chats. A few are building an operating system for their brain. Which one are you?

THIS GUY WIRED A RACK OF $130 TESLA M40 CARDS INTO A HOME CLUSTER THAT KILLS A $412 A MONTH AI STACK ON $4 OF ELECTRICITY. The whole thing lives in a hallway cupboard in Sydney. A NETGEAR switch, a row of cheap boxes bolted into an open frame, blue cables zip-tied in neat runs, one red rocker switch and a fan that never gets loud. Cooper is 27 and had never wired anything before this. What set him off was his card statement. $412 a month going out to AI subscriptions. Claude Max, ChatGPT Pro, Cursor, two transcribers he forgot he signed up for. $4,944 a year to rent compute that now fits in a cupboard. So he built the cupboard. Tesla M40 cards off eBay at $130 each, 24GB apiece, dropped into old machines and wired into the switch. Three commands and Ollama was running the same class of models the subscriptions were renting him. Free after electricity, about $4 a month. The real move is not the rack. It's running three brains instead of one, each pointed at what it does best. Claude Sonnet does the heavy thinking. Claude Haiku takes the dumb repetitive stuff like tagging, twelve times cheaper. The local box handles anything private that should never leave the house. You don't pay a surgeon to change a lightbulb. Then he stopped pushing the buttons himself. The rack works while he sleeps. It sorts the inbox into reply, read, trash and drafts the replies by morning. It reads everything he saves and files it. It watches a list and pings his phone only when something actually moves. Then people around him wanted one. So he builds the rack for them. $940 a build, parts run him $280, one night of wiring. Nine builds since March. $5,940 in his pocket, plus the $4,944 a year he stopped paying himself. Everyone else still rents what hums in his cupboard for $4 of power a month. Bookmark before the market notices.

@0xCodila makes sense but depends hard on what task you're actually solving

Anthropic Engineer Andrej Karpathy: "Stop training from scratch. Take what the model already learned - adapt it - ship it At OpenAI we replaced months of work with a few lines of code. Better results every time " the Karpathy formula for working with AI: step 1 → stop building from zero. the model already learned the hard part - just point it at your task step 2 → stop adding complexity, the simplest architecture beat everything in 2014 - simpler always wins step 3 → train only the last layer and freeze everything else - minutes instead of months. better results he taught this at Stanford 10 years ago - then applied it at OpenAI, Tesla, and Anthropic the advice never changed - because it never stopped working watch - bookmark, then read article below ↓

@chewadot i gonna start buiding obsidian vault rn bro

ANTHROPIC'S LEAD ENGINEER ACCIDENTALLY LEAKED HIS PERSONAL OBSIDIAN VAULT. INSIDE - NO CODE, NO PROMPTS. JUST A SCHEMATIC OF HIS OWN MIND, BUILT LIKE A NEURAL NETWORK 8,893 nodes. 4,729 links. A $10-a-month app opens Obsidian. 21 inputs, 10+ hidden layers, ReLU activation. First layer 64 neurons, then 37, then 22, all the way to the output. Thousands of connections firing in real time this isn't a concept diagram from a blog - it's a living brain that actually runs decisions inside the company. 9,000 documents, each its own knowledge space, all interconnected he makes around $2M a year for putting markdown files into the right folders. The company building the best AI in the world manages its internal knowledge with the same app a college freshman uses for lecture notes. hasn't written a single line of infrastructure for any of it. three years of discipline and one open Obsidian tab you're reading this on a device where you could open the same Obsidian tonight and start building your own vault

@noisyb0y1 boris is the goat

CLAUDE CODE CREATOR BORIS CHERNY WENT FROM 0 → $50M IN 12 MONTHS "90% of code I no longer write by hand. I built the right context and forgot about work." 30 minutes - full step by step guide on Context Engineering from the lead engineer retrieval → compression → memory → routing → verification - works without you 24/7 fixes errors on its own, writes new lines of code, builds the entire project skeleton bookmark & watch the video below - and the agent works without you

@yurshevv yes, this is simple asf and saves you a ton of money

@raidenfomo turning $310 of enterprise scrap metal into a local ai powerhouse takes just one evening

THIS GUY TURNED $310 OF RETIRED DATACENTER HARDWARE INTO A BOX THAT KILLED HIS $412/MONTH AI SUBSCRIPTION STACK IN ONE EVENING. A firm in Katowice was scrapping its server room. The repair guy they called to haul it out is 34, has a shop across the street. He kept one Dell for himself. $180 off the invoice. Everyone saw scrap metal. He saw a slot for a datacenter GPU. He slid one card into that Dell: an NVIDIA Tesla M40. $130 on eBay, 24GB of VRAM. The same 24GB you get in a $2,000 card today. NVIDIA sold it to server farms for $3,500 in 2015. The farms dumped it. Nobody wanted it back. Three commands to install Ollama and pull a 27B model. One evening. The box runs it fine. Pause at 0:24 on the front badge, that is a PowerEdge R720xd, gear that sold new for around $8,000. Now it works while he sleeps. Sorts his inbox into three piles at night and drafts the replies. Boils every article he saves down to a summary. Messages his phone when a price he watches moves. He checks it once a day. His old bill: Claude Max, ChatGPT Pro, Cursor, two transcribers he forgot he had. $412 a month. $4,944 a year. New bill: $310 once, $19 a month for electricity. The hardware costs less than one month of the old stack. Datacenters retired this box in 2019. In 2026 it is wanted more than ever. Save this before the market notices.

@eng_khairallah1 this is insane if his actually running 40 agents in parallel rn

Ex-Google AI engineer: "We will be running hundreds of agents working for us in the near future. I'm running 40 agents in parallel. I have built a $50,000 marketing team for just $500 using AI." In 15 minutes she shows exactly how to build an agent that improves itself. Most people are still doing all of this by hand. Worth more than a $500 AI course on the internet. Watch today, then read the article below on the same workflow.

@undefinedKi this video is literally pure gold

If you're using Fable 5 on default settings, you're literally burning money Fable 5 is the most powerful model out there, but it's expensive and your free window closes July 7. Most people run it wrong: max effort, planning and executing everything on it. That's how you hit your limit in days. Here's how to cut Fable's cost by up to 80% without losing what makes it great: 1. Drop the effort level. This is the big one. On DeepSweet, max effort costs ~$22 per task. Low effort costs $3.76, over 80% cheaper, and still beats Opus 4.8 at max. Most tasks don't need high. For web design or simpler work, use low or medium. Just type /effort and set it. 2. Make Fable the architect, not the laborer. Don't let it plan AND execute everything. Have it write the plan, then hand the actual building to a cheaper model (Opus, Sonnet, or GPT via the Codex plugin). Fable is smart enough to say which model fits each part. 3. Add a token-cutting skill. Tools like Ponytail give Claude guidelines to write less code for the same result. On Fable at medium, it came out ~22% cheaper. On an expensive model, a 20% cut is real money. 4. Research with a cheaper model, plan with Fable. Deep research spawns 100+ sub-agents, you do NOT want those running on Fable. Use Opus or Sonnet with /deep-research to gather context, then hand it to Fable to build the plan. 5. Use advisor mode. Set your model to Opus, then /advisor fable. Now Fable acts as the brain directing a cheaper executor, high-level thinking from Fable, cheap execution underneath. The pattern: Fable is a specialist for hard thinking, not a workhorse for everything. Use it that way and it lasts. Bookmark this.

@kobaHUB yo this is gem

$10,000 vs. $0. One afternoon vs. weeks of back-and-forth with an agency. Here's how Claude builds you a full website for free. Agencies charge $10K for a website. Here's how to build your own with Claude for free. Write a clear, detailed brief - type, style, sections. Specific beats vague every time. Ask for a complete, working HTML file with real content, not placeholders. Save as index.html and open it in your browser. Describe your visual style in plain English - colors, fonts, spacing. Claude rewrites the CSS. Add interactivity: smooth scroll, mobile menu, form validation. One feature per prompt. Test on mobile and desktop. Report specific issues ("heading wraps to 3 lines on iPhone") for fast fixes. Deploy free on Netlify - drag your folder, get a live URL in 60 seconds. Add a custom domain anytime. No CSS or JS knowledge needed. Just clear prompts and iteration.

@aiseomastery yea this is crazy

@raidenfomo A $340 scrap server replacing $412 a month in subscriptions is such a wild flex.

THIS GUY BOUGHT A RETIRED $340 DELL SERVER TO KILL HIS OWN $412/MONTH AI SUBSCRIPTION STACK AND SAVE $4,944 A YEAR. A 29 year old in Cleveland. He works in a shop that wipes and flips decommissioned corporate servers. Companies pay to get rid of them. One day he did the math on his own credit card. $412 a month in AI subscriptions - Claude Max, ChatGPT Pro, Cursor, Perplexity, two transcribers he forgot he signed up for. Then he looked at the pallet next to him and realized the models behind all of it run on the exact box a corporate IT department pays to throw out. He grabbed one. Dell PowerEdge T550, $340. He loads Llama 70B on it for the local work, points Claude at the jobs that need real thinking, and closes the lid. Three commands, ten minutes. Now it runs while he sleeps. It sorts his inbox into reply, FYI and trash, and writes the replies. It reads every article he saves to a folder and links it to the old ones. Nothing ever leaves the house. He cancelled every subscription but one. $340, once. $412 a month, gone. $4,944 a year he was renting compute that fits on a box he pulled off a scrap pile. The offices that paid to junk these are still renting the same thing back at $412 a month. Save this before everyone catches on.

@AnatoliKopadze this is gem bro

Anthropic Managed Agents Lead: "Over 90% of Anthropic engineers already build with self-improving loops. My agentic loops run for hours and cost almost nothing." In 40 minutes she builds an AI agent from scratch and shows how engineers at Anthropic use them daily. This video is worth more than any $600 course on agents. Watch it today, then read the guide on building loops below.

@noisyb0y1 helpful share brother

OPENAI DEVELOPER MAKING $1.7M/YEAR JUST LEAKED THE GUIDE ON HOW THEY BUILD AI SYSTEMS - AND GOT ARRESTED WITHIN THE HOUR regular AI responds and forgets - a loop responds, learns and gets 4x better every time Plan → Execute → Evaluate → Memory → Plan five layers, one cycle, accumulates knowledge over time every completed cycle produces better data - better data leads to better decisions - better decisions lead to stronger execution the engineer no longer optimizes prompts - the engineer designs execution 5 results of Loop Engineering systems: higher output quality - fewer errors - lower cost - better safety - continuous improvement the developer defines the goals and the system does everything else this is the new standard for building reliable, scalable and intelligent agentic systems

@cyrilXBT i need to try this

KARPATHY'S SECOND BRAIN HAS ONE RULE. He does not write it. He drops raw sources into one folder. Articles, papers, repos, images. Nothing cleaned. Nothing sorted by hand. The model reads it and builds a wiki of linked markdown. Summarizes every source. Groups the ideas. Backlinks everything. Once the wiki is big enough, he questions it like a researcher. The model runs health checks on its own work. Finds the gaps. Suggests what to look into next. RAW belongs to him. WIKI belongs to the model. Every answer gets filed back, so the brain compounds instead of resetting. Bookmark this. Follow @cyrilXBT