Ochob

@0xOrionVega I'm at 180 trades. Now i know why I keep changing systems

Write down what you think is a random sequence of 100 coin flips. Persi Diaconis will look at your paper and know within seconds you made it up. Real random has streaks. Yours doesn't. And that's why the chart in front of you doesn't mean what you think it means. Diaconis is a Stanford math professor and a former professional magician. He has spent forty years running this experiment on his students. He tells human sequences from real random ones with about 95 percent accuracy. Real randomness writes seven heads in a row without hesitating. Nobody in the class ever writes that. The brain refuses to let it happen. His free lecture on YouTube is called "Does Anything Happen at Random?" The short answer he gives is: much less than you think. Not coins. Not shuffled decks. Not markets. A trader who sees a losing week and thinks the system is broken is running the same failure as those students. Real randomness produces losing weeks constantly. The feeling that something must be wrong is the same instinct Diaconis catches every semester. The market is not pure noise either. Order flow, positioning, forced flows, rebalances all leave patterns too small for the eye and obvious to the math. Most of what feels like a hunch is the brain inventing structure. Most of what feels like noise already has structure. The math also gives you a number. To statistically prove a 55 percent edge is real, you need about 400 trades. For a 52 percent edge, roughly 2,500. For a 51 percent edge, over 10,000. Almost every retail trader quits, changes systems, or decides their model is broken long before reaching that sample size. They exit the experiment before the answer arrives. The math has been free for 300 years. Trusting it long enough to see the answer is what almost nobody does.

@PoolCleaner6 But everyone knows him :)

@0chob Warren Buffett kind a meme for me

Warren Buffett reads Howard Marks first. In October 2008, Marks told his readers to buy while banks were collapsing. He deployed seven billion dollars in ten weeks and made twenty percent a year on the trade. He has never predicted a market crash in his life. Marks does not predict anything. That is the point. He founded Oaktree Capital in 1995 on one thesis: nobody knows where the market is going. What you can know is how much risk everyone else is taking, and how they are pricing it. When risk is cheap, you sell. When it is priced as if the world is ending, you buy. In the fall of 2008, everyone was pricing risk as if the world was ending. Lehman had collapsed. AIG was being nationalized. High-yield bonds were trading at 40 cents on the dollar. Marks published a memo titled "The Limits to Negativism" and told his readers: this is what you were waiting for. Oaktree deployed seven billion dollars in the fourth quarter. By 2011, those distressed positions had returned twenty percent a year, net of fees. Marks does not treat this as luck. He treats it as the only way distressed investing works. He gave the framework away in a lecture at the Fundación Juan March in Madrid in May 2019. The whole method came down to one distinction he had been making for forty years: "There is a difference between buying something good and buying well." A great company at a bad price is buying something good and paying too much. A mediocre company at a great price is buying well and getting paid. Distressed investing lives entirely in the second category. So does most of the money in Oaktree's history. Marks has written memos every quarter since 1990. He does not predict. He does not forecast. He does not have a view on the Fed. He measures where risk is being priced and acts when the market is wrong. Then he waits. He has said the same thing five thousand times in memos, letters, and lectures: "You can't predict. You can prepare." He built a two-hundred-billion-dollar firm on that one sentence. The Madrid lecture is on YouTube. The memos going back to 1990 are on Oaktree's website. Both are free. Buffett reads them first. Everyone else waits until they need them.

@crimson_moti She is so real

A GIRL WITH PINK HAIR IS PADDLEBOARDING IN THE OCEAN AND SHE IS WHY YOUR AIRBNB LISTING IS DEAD. She sits on a turquoise board. Pink hair in the sun. Black bikini. She runs her hand through her hair, stands up with the paddle. The whole clip looks like a $3,000 production day in the Caribbean. It wasn't. She does not exist. And the person who generated her charges less than the cost of parking at that beach. Pause at 0:04. Look at the muscle definition on her back. Look at the water reflections on her skin. That is not a vacation video. That is a listing asset. And this article breaks down how one operator turned this kind of content into $12,900 a month staging Airbnb homes without ever booking a photographer. HERE'S WHAT MOST PEOPLE WILL SCROLL PAST. The operator runs a full book of short-term-rental hosts. Six tools. Eleven hours a week. Every host knows their photos are the reason bookings stalled but none of them want to reshoot. So the operator generates the entire visual identity from scratch. Staging. Lifestyle shots. Property tours. One person with one pipeline generating clips like the one above. Pause at 0:07. The wide shot from behind. Turquoise board on calm water. Put that in a beach condo listing next to every other host's iPhone photos from 2021 and the click-through difference is not even close. The modeled ceiling is $12,900 a month. Every US market has a stack of properties with dead photos and every one of them is one better image away from a booking they are currently losing. This girl with pink hair on a paddleboard is that image. And the person who made her doesn't own a camera. Watch the clip. Then open your nearest Airbnb listing. That's the gap.

@kursormaxx From $800/mo burn to local inference - solid pivot

Giorgi Kavtaradze runs sports betting arb bots out of a Vera district apartment in Tbilisi. Half in Polymarket, half in regulated Georgian books. Started on a MacBook Air M2, then bolted a used Razer Core X + RX 6800 XT to it for the model training. His OpenAI API bill hit $400 in March when the NCAA bracket ran. RunPod on-demand for backtests added $180. Bloomberg Terminal-style Sentix Pro sub, Kalshi Pro, TradingView Premium — another $220/mo just to watch the tape. Total burn crossed $800 last month. Cancelled everything in April. Runs Qwen 2.5-72B locally on the 6800 XT for reading news feeds, whisper on prop firm podcasts, a fine-tuned Llama 3.3 for player-injury sentiment scoring. The eGPU pulls 320W under load, Georgian electricity is 0.11 GEL/kWh. The bill scaled. The box doesn't.

@kursormaxx Mispricing shows up loudest when everyone else is running

@0chob The biggest investing edge isn't predicting the future. It's staying prepared for outcomes everyone else is pricing incorrectly

@Primee32 And it fits on a napkin

@0chob "buying something good vs buying well" is doing more work than most entire risk frameworks

@0xOrionVega That one costs everybody. Mine was closer to 5

@0chob The buying well vs buying good thing was 3 years of my portfolio in one sentence

@velesxbt Yeah. The article did the heavy lifting. This is just marks putting a face on it

@0chob The 2008 memo is layer 5 in 8 pages

@Primee32 12 frames, one strike - that's how you nail sakuga. Want me to post this?

THE SHOT THAT MAKES A LOCKED SWORD STANCE FEEL LIKE A REAL SAKUGA STRIKE One stance under a blood-red moon. Twelve frames later, three creatures are on the ground and the blade hasn't stopped glowing. Most creators lock a beautiful character sheet, generate one cool pose, and call it action. Real sakuga-style motion means that same pose has to survive twelve consecutive frames of violence without the silhouette ever going mushy. Here's the workflow 1. Build the character concept and lock it in GPT Image 2 first — pale windswept hair, dark flowing coat, glowing blue energy blade, calm low stance before the strike. Front, side, and grip close-up, with detail callouts on the coat silhouette and the blade's exact glow color 2. Storyboard the strike as twelve named beats instead of one action line — low moon stance, ground threat, wrist snap, impact profile, low block, rear pressure, high lane, ground evidence, frontal impact, creature POV, side release, low aftermath. Each panel gets its own label, not just a number 3. Strip the palette down to two tones and let them carry the whole scene — a blown-out red moon and near-black silhouettes, with the blade's blue glow as the only color allowed to cut through both 4. Keep every enemy as pure silhouette, never detailed — this isn't laziness, it's what keeps all visual attention locked on the one character sheet that actually needs to hold up 5. Let the impact and aftermath panels do the identity-proofing — motion blur and slash streaks are exactly where a rushed design falls apart, so give them their own dedicated frames instead of compressing the strike into one blurry beat 6. Animate the full sequence in Seedance 2.0 straight from the twelve-panel board, with the coat silhouette and blade color repeated explicitly in every shot Why this works - Spending all twelve panels on a single strike instead of a whole fight is what makes the motion read as fluid, not chopped — every frame is a phase of the same action, not a new idea - Naming every panel instead of numbering it forces a real decision about what happens in that specific frame, not a vague "action continues" - A two-tone palette with one glowing exception makes the blade the only thing your eye can track across all twelve panels, even at full speed - Keeping every enemy in pure silhouette removes any competition for attention — the audience is never confused about whose consistency actually matters Use cases: ⁃ Single-hero sakuga-style action sequences ⁃ Silhouette-driven fight scenes against a strong color backdrop ⁃ Any strike or impact moment that needs to read clearly at high speed ⁃ Dark fantasy or shonen-style openers built around one signature weapon The character sheet answers what the stance looks like standing still. Twelve frames of one strike are what prove the blade was never standing still at all.

@velesxbt Medallion still closed to outsiders?

Jim Simons wrote a theorem in 1974 that now underpins string theory and quantum computing. He also built the highest-returning hedge fund in history. He is the only person who ever did both. He died in 2024. His fund is still running, and nobody outside it knows how. The theorem is called Chern-Simons theory. Simons co-authored it with Shiing-Shen Chern as pure geometry, with no application in mind. Physicists needed it thirty years later to describe the fundamental structure of quantum fields. Microsoft is now using it to build quantum computers. It was Simons' side project. His day job was Stony Brook. He chaired the math department. In 1976 he won the Veblen Prize in geometry. He was a legend to mathematicians and unknown to everyone else. He did not need Wall Street. He was already famous inside a room of people who don't care about money. He left in 1978 anyway. The first years were rough. He and his partners tried trading on intuition. It didn't work. By 1989 Renaissance was underwater and the partners were arguing about whether to shut it down. Simons considered walking away. Instead he made one decision: no more intuition. Every trade Renaissance would ever make from that point on would be automated, driven by patterns in data, executed by code. No human would ever again overrule the system. The Medallion Fund launched with that rule locked in. From 1988 to 2018, it returned 66% a year before fees and 39% a year after. The highest sustained return in financial history. Simons hired mathematicians, physicists, and NSA cryptographers. He never hired from Wall Street. He said Wall Street traders had spent their careers learning to trust their gut, and gut was exactly what Renaissance had banned in 1989. He rarely gave interviews. When he did, he sounded almost bored. He chain-smoked through every meeting for thirty years and refused to wear socks. Nobody at the firm ever asked him to stop. Asked once why he hired scientists instead of bankers, he answered in five words: "We look for anomalies that maybe were repetitive." That was the whole game. The math was not exotic. The tools were widely known. But everyone who tried to copy Simons missed the one thing Renaissance figured out in 1989: the human is the bug. Every thousand dollars invested in Medallion in 1988 grew to over twenty million dollars by 2018 after fees. The same thousand dollars in the S&P 500 grew to twenty-three thousand. Simons gave most of his fortune away before he died. Math education. Autism research. Public science. Quantum computing — the field his 1974 theorem now powers. Your instincts are the fee you pay to feel like you know what you are doing. Simons stopped paying that fee in 1989. Everything else that happened was math.

@crimson_moti Wild how far synthetic realism has come already

THIS GIRL IS LYING ON A WHITE PILLOW WITH MESSY HAIR AND FRECKLES AND SOMEONE CHARGES $800 TO BUILD HER. She wakes up. Messy blonde hair on a white pillow. No makeup. Wrinkled white t-shirt. She picks up the phone, looks into the camera half asleep, and asks "can you tell?" Then the camera pushes in. Close-up. Freckles. Pores. The redness around her nose. The kind of skin detail you only see in a mirror first thing in the morning. Every single pixel of it is AI. Pause at 0:05. Look at the skin. That's not a filter. That's not a texture pack someone dragged onto a face. Those are individual pores and freckles that the system generated because the negative prompt was set up before the render. Most AI faces can't survive a close-up like that. This one leans into it. I'm not saying AI faces are new. Calm down. We've all seen the perfect symmetrical doll faces that fool nobody. But this is not that. This is a "just woke up" selfie with bed-head and skin imperfections and the look on her face when she hasn't had coffee yet. That's the part that should stop you. Nobody told the AI to add freckles. The system did it because the skin texture pass was set up correctly before the face was ever animated. HERE'S WHAT MOST PEOPLE MISS ABOUT WHY THIS CLIP WORKS. The creators who generate perfect AI faces wonder why nobody engages. It's because perfect is the tell. The human eye doesn't trust symmetry. It doesn't trust flawless skin on a morning selfie. It looks for the things that shouldn't be there if the face was generated: the freckle, the pore, the messy strand of hair covering one eye, the slight redness on the nose. This clip has all of them. That's not luck. That's a negative prompt node running 40+ exclusion terms that strip the "AI look" before the face is ever rendered. I build AI personas and the mechanic underneath this clip is the one thing I tell every creator who asks why their content looks fake. You didn't break the face. You skipped the skin texture pass. You sent a clean symmetrical render straight into animation and the audience clocked it as AI in the first second. This girl has freckles on a pillow because someone ran the negative prompt correctly and refused to animate until the skin passed the close-up test. Pause at 0:08. She touches her chest and says "this is AI." By that point most people have already forgotten they're watching a generated face. That's the whole game. The freckles did the work in the first five seconds and after that the brain stops questioning. The creators adding imperfections on purpose are the ones whose personas survive past week one and clear $800 per character. Not because imperfection is a style choice. Because imperfection is what makes a human brain stop analyzing and start believing. And the moment someone believes the face they follow it and they pay for it. Watch it twice. Count the freckles at 0:05. That's the pipeline nobody talks about.

@0xFramez This is a goldmine for AI engineering. Thanks for sharing - bookmarked!

Andrew Ng just dropped a 3-hour course on how to become an AI Engineer in 2026. 6 months ago I would have said learning AI engineering takes a year. You need to grind through courses, read papers, and break things in production before you get it. This course changed my mind. In 3 hours Ng takes you from knowing nothing about agents to building a full working app on camera. Not slides. Not theory. Real code, real product, real deployment. I have paid for 4 agent courses this year. Every single one of them is in this video, for free. The prompting section alone at 23:38 is better than most paid courses I have seen. The reason is simple. Most course creators learned AI from Twitter threads 18 months ago. Andrew Ng spent 20 years at Stanford, built Google Brain, ran AI at Baidu, and created Coursera. He is not teaching what he read last week. He is teaching what he built his entire career on. Here is what separates this from everything else out there. He does not just show you how to use a framework. He shows you how to think about agentic systems so you can build with any framework, any model, any stack. That is the difference between someone who follows tutorials and someone who actually ships AI products. Everyone I sent this to had the same reaction. "Why did I pay for courses when this exists." 3 hours. Zero to a working AI app: · 00:00 – Build agentic AI systems from scratch · 04:25 – Where AI engineering is actually headed · 23:38 – The full AI prompting course · 2:52:17 – Build a working app with AI in 30 minutes Watch it today, then go deeper with the full guide on building AI agents below.Andrew Ng just dropped a 3-hour course on how to become an AI Engineer in 2026.

@0xOrionVega Worked until the last week. That's the flex

For years, one man argued with Warren Buffett about a candy company. When Buffett finally caved, that single trade rewrote every rule of investing. His name was Charlie Munger. He died in 2023 at 99. Almost nobody has learned what he tried to teach. Before 1972, Buffett bought garbage. He called them cigar butts. Cheap, dying companies with one last puff left in them. It worked. Buffett was rich by 40. He was going to stay moderately rich forever. Then Munger arrived. A one-eyed lawyer from Omaha who thought Buffett was doing it wrong. Munger did not want cheap. He wanted great. The philosophy fit on a business card: "It is far better to buy a wonderful company at a fair price than a fair company at a wonderful price." It took him years to sell it to Buffett. In 1972, Munger picked the fight worth having. See's Candy was for sale for $25 million. Book value was $8 million. Buffett said no. Munger said yes. Munger won. See's returned that $25 million more than eighty times over. Buffett stopped hunting cigar butts. He bought Coca-Cola in 1988. American Express in the 1990s. Apple in 2016. Every trade was Munger's playbook, executed in Buffett's name. The final ledger: over a hundred billion dollars generated by one argument won in 1972. Munger's other teachings were short. Invert every problem. Know your circle of competence. Follow the incentives. Sit on your ass and wait. Bet big when you finally act. Do not be stupid. He said them thousands of times, for free, at every Berkshire meeting for sixty years. He explained the whole game in one sentence: "It is remarkable how much long-term advantage people like us have gotten by trying to be consistently not stupid, instead of trying to be very intelligent." In late 2023, three months before his death, Munger gave his final interview. He was 99. He said nothing new. He never had to. The material had been correct the first time. He died a month before his hundredth birthday. He worked until the last week. The lesson is still free. Wonderful businesses. Fair prices. Long waits. Big bets. Not stupid. Almost nobody applies it.

@Leonholdss @Polymarket France will prove power beats possession tonight

Tonight isn’t just France vs Spain. It’s power vs control. Physicality vs possession. One team can punish you in a single moment. The other can slowly take the game away from you. @Polymarket has France at 60% and Spain at 40%. One match away from the World Cup final. Who are you backing tonight? 🇫🇷🇪🇸 #WorldCup2026 #France #Spain #Football @Polymarket

@misat0x Haha, exactly

@0chob okay now i need proof he’s real too😬

MrBeast reportedly spends tens of millions every year making his videos feel native in other languages. This guy just showed the small-creator version of that strategy: take one video dub it into 40+ languages keep the original voice sync the lips The demo is impressive. But the interesting part is not AI dubbing. It is what happens after the dub. Most creators treat a finished TikTok as one post. The smarter move is to treat it as a source asset. One useful English tutorial can become a version for Mexico and Colombia. Then Brazil. Then another market where the same problem exists, but the content still feels imported. Not by throwing subtitles on it. By rebuilding the hook, slang, examples, captions, and on-screen text until a viewer feels: this was made for me. That is the opportunity. The new creator advantage is not making more videos. It is giving the good ones more than one chance to win. More audience. More feedback. More ways to make money. I mapped out the full workflow: which market to test first, how to localize the script, dub it with ElevenLabs, fix the visuals, and turn the winners into a repeatable system. The tools just made global distribution cheap. Making content that actually feels local is still hard. Full article below 👇

@Psalteric Great breakdown - infrastructure catching up is the real story.

OpenAI, xAI, and Meta dropped frontier-level AI models on the same day — and Cloudflare responded by making AI agents cheaper to run. But the real story isn’t another benchmark leaderboard. It’s how quickly frontier intelligence is turning into infrastructure. Here’s what happened: – OpenAI released the GPT-5.6 series publicly, pushing coding performance forward again – xAI launched a new model designed to compete at the frontier — with benchmarks placing it alongside the leading labs – Meta released Spark 1.1, an agentic coding model with serious capabilities at a fraction of the usual price – Then Cloudflare reset its weekly rate limits for all users — giving developers more room to actually build with these models The takeaway: frontier AI is no longer just getting smarter. It’s getting cheaper, more accessible, and easier to deploy at scale. Three labs shipped. Infrastructure providers reacted. Developers got more leverage overnight. The AI race just shifted from “Who has the best model?” to “Who can turn these models into working products fastest?”

@0xFramez OpenAI's going full platform mode. Worth the watch.

OpenAI CEO Sam Altman just dropped the full DevDay 2025 keynote. Apps inside ChatGPT, AgentKit, Codex, and the new Apps SDK. Everything developers need to build on top of OpenAI, in 50 minutes: · 00:00 – Sam Altman opens DevDay 2025 · 02:00 – Two announcements that change everything · 04:00 – Apps SDK: monetization, login, actions · 06:00 – Coursera running inside ChatGPT 5 · 10:00 – Canva integration, live on stage · 12:00 – Zillow real-time search inside ChatGPT · 16:00 – Every agent type OpenAI is building · 18:00 – Agents plus Connectors, explained · 20:00 – AgentKit launches today · 22:00 – Building a multi-agent workflow live · 30:00 – Sam Altman on the future of agents · 34:00 – Codex: building an MCP server on stage · 44:00 – Live agent controlling a camera in real time · 48:00 – What comes next for developers Most people read the announcements on Twitter. This 50-minute keynote is the actual playbook