Kirill 🔰 MBA

An ex-OpenAI engineer walked up to me at a meetup in SF I was showing my Polymarket terminal. He looked at the screen for ten seconds and said one sentence. "You're trading blind. The data is sitting in the open and you're writing prompts" I didn't understand. What data. He took my laptop. Opened one repository. github.com/warproxxx/poly… 86 million trades. Every wallet. Every entry. Every exit. The entire Polymarket history since day one. "At OpenAI models don't guess. They read. Connect Claude Code to this dataset and say - find every wallet with a win rate above 70% and more than 100 trades" I asked - why Claude and not GPT? He looked at me like I was an idiot. "Because Claude Code connects to the repo directly. It reads the entire codebase. It's not a chatbox. It's a runtime" That evening I connected it. Claude pulled 47 wallets in 4 minutes. Average profit: $214K. Hold time: 7 hours. 91% close their position BEFORE resolution. Top wallets capture 86% of the move and cut losers at 12%. Everyone else - 58% of profit and hold losers to 41%. Same exact entries. The exits make it a completely different game. "Now connect the scanner" github.com/Polymarket/pol… Three commands - the bot sees 500+ markets in real time. No API key. Read-only. Claude built the scoring in 20 minutes: Gap between price and model > 7 cents. Book depth > $500. Resolution in 4-48 hours. 93% of markets get killed instantly. Only the fat ones survive. "Now this is the important part" - he sent me an article where a guy built a full bot over a weekend from these same repos Copytrade here: @lunar" target="_blank" rel="nofollow noopener">kreo.app/@lunar Three exit triggers: Target 85% of expected move. Volume spike x3 - smart money leaving. 24 hours of silence - thesis is dead. I copied the whole stack. VPS $5. Claude $20. Total $25 a month. No team. No office. No Bloomberg. 16 days. 187 trades. 71% win rate. $800 seed. +$8,700. I sent him my screen. He replied a day later. "You just replicated for $25 what cost us six months and 11 people" I said - thanks for the tip. "Delete this chat" Too late.

🚨JUST IN The Federal Reserve will inject $40.46B into the markets over the next few weeks. QE is back. The money printer is officially on again. This is EXTREMELY bullish for risk assets! 🚀

As the West Asia conflict enters a new phase, Chinese cargo aircraft landed at Quetta Airbase in Pakistan yesterday, carrying a large quantity of arms and ammunition, including anti-aircraft missiles. These were later transported to Iran via the Pakistan–Iran land border route.

We just OCR'd 27,000 arxiv papers into Markdown using an open 5B model, 16 parallel HF Jobs on L40S GPUs, and a mounted bucket. Total cost: $850 Total time: ~29 hours Jobs that crashed: 0 This now powers "Chat with your paper" on hf.co/papers

🚨 Google acaba de liberar una de sus herramientas internas más potentes. Ha lanzado Magika, un sistema que detecta el contenido real de cualquier archivo, aunque esté disfrazado. Por ejemplo, malware que parece un PDF o scripts ocultos como imágenes. Es gratis y open-source 👇

lmfao westerner tax from @Zai_org is crazy The same "Max" plan served in China that costs $68 USD (469 RMB) is $160 USD for westerners. Irony is, if you have WeChat or Alipay, you can just buy the Chinese plan and still have API access. So it's not even GEOLOCK related costs. It's just capitalising on the gold rush right now for subscription-based OAuth integrations. Diabolical is an understatement tbh

🇸🇦🇮🇷🇺🇸 SAUDI ARABIA Royal source: "US Must Launch Ground Invasion of IRAN Like Saddam's Fall — Avoid Reza Pahlavi" “There is no solution other than a ground military entry by the US into Iran and a repeat of the overthrow of Saddam Hussein's regime in 2003. But this time we need to find a successful, acceptable, and secular alternative government on the Iranian street. It would be better if it wasn't Reza Pahlavi. There is no other solution. They should treat this like the Saddam Hussein regime, and recreate the Bush administration's experience. The war has not achieved its results, and Iran is continuing with its nuclear program, and has even become more aggressive towards us, towards you [Israel], and towards the US and its interests in the region and the world.”

I found a GitHub repo that reads doctor handwriting. It's called Chandra. It handles forms, tables, math equations and cursive documents that have defeated OCR for 30 years. One command. Full layout preserved. → Reads handwriting, cursive, and messy print with full accuracy → Reconstructs merged table cells including colspan and rowspan → Renders inline math and block equations as LaTeX automatically → Extracts checkboxes, radio buttons, and form fields with their values → Outputs Markdown, HTML, or JSON with bounding box coordinates for every element → Supports 40+ languages out of the box → Runs locally via HuggingFace or on a vLLM server for production throughput Tested on 10K financial filings, medical forms, LA Times newspapers, and college worksheets. It beats every baseline on the OLMOCr benchmark. 100% Open Source. Apache 2.0. 4.8K stars. Repo: github.com/datalab-to/cha…

Gemma 4 analyzes the video. Generates key questions. Calls Falcon Perception. "Find all the people." 156 found. "Detect only white cars." 8 found. A 26B model is running agentic multi-QA vision orchestration. The models are running locally on a MacBook with MLX. No API.

THIS TOOL SHOWS YOU EXACTLY WHERE YOUR CLAUDE CODE TOKENS ARE GOING this guy was spending $200+ a day on Claude Code with zero visibility into what was eating the tokens. so he built a terminal dashboard that reads your session transcripts and classifies every single turn into 13 categories with no LLM calls. what it shows you: > cost by task type (coding, debugging, exploration, brainstorming, etc) > cost by project, model, tool, and MCP server > daily activity chart > AND its interactive: arrow keys to switch between today, week, and month 56% of his spend was "conversation" where Claude is just responding with no tool use and the actual coding (edits and writes) was only 21%. more than half the money was going to Claude thinking out loud instead of actually writing code. one line install: npx codeburn free AND open source.

China Exports Growth Slows Sharply: Imports Surge China exports in dollar terms rose 2.5% y/y in March, far below expectations for an 8.6% increase. Imports jumped 27.8% y/y, significantly beating expectations of a 13.9% rise.

This is it, a Gemma 4 31B distilled on Claude Opus 4.6 reasoning has been released! Gemopus-4-31B-it-GGUF It was added 5min ago, ggufs are being added as we speak. Jackrong & @KyleHessling1 are on a generational run 🔥 huggingface.co/Jackrong/Gemop…

🇲🇾 Malaysia just seized two tankers caught doing illegal ship-to-ship oil transfers in the Malacca Strait, the same waters where Russian and Iranian crude quietly changes hands out of sight. The U.S. blockades Hormuz. Malaysia cleans up the backdoor. The shadow fleet is having a very bad week.

Google quietly open sourced a time-series AI that predicts anything. Sales trends. Market prices. User traffic. Energy demand. Crypto volatility. It's called TimesFM. Pre-trained on 100B real-world data points. Zero-shot forecasting with no fine-tuning. Outperforms supervised models trained on your specific data. Runs locally. Free. Apache license. Most people are focused on language models. The quietly powerful ones are learning to predict the future.

🚀 New TTS Model: MOSS-TTS-Nano, a Real-time TTS that runs fully on CPU! 💡I'm always looking for good TTS models to run on an average PC (w/o a GPU) and this is one of them. OpenMOSS is a new 0.1B multilingual model: • No GPU needed, pure CPU inference • Real-time speech generation + streaming • 20 languages supported • 48 kHz stereo quality • Ultra-lightweight & easy local/web deployment Huge value for edge devices, local demos, web apps, and lightweight products where GPU isn’t an option. Part of the MOSS-TTS family 👏

This 1 hour Yale lecture will teach you more about options trading & the exact models Hedge Funds use than most people learn in their entire careers on Wall Street. Bookmark this & watch, no matter what. It’ll be the most productive start for your week. Then read article below.

🚨 WARNING: SOMETHING VERY UNUSUAL IS HAPPENING RIGHT NOW Insiders are buying Gold options at $20,000 for December 2026. Gold is sitting at $4,700 today. Yes, they expect the gold price TO PUMP OVER 300% by the end of this year. And if you think this is just reckless gambling... YOU'RE COMPLETELY WRONG. Let me break it down simply: This position did NOT appear at the top. It started building AFTER gold pushed above $5,600 earlier. Right before getting slammed by the biggest one-day drop in decades. That’s the detail almost everyone ignores. Retail panic sold. These buyers kept stacking. Even as gold dropped back towards $4,200. Now the position has grown to around 11,000 contracts. That’s roughly 1.1 MILLION ounces. Roughly $5.17 BILLION at current prices. Roughly $17.5 BILLION at the $20,000 strike. That is NOT normal activity. That is a tail-risk bet on a complete system repricing. Now connect the dots. Mainstream bank targets for 2026 sit around $6,100–$6,300. This trade only starts to matter around $20,000. That tells you everything. This is NOT positioning for a standard bull market. This is positioning for a monetary shock, a systemic crisis, or a market break big enough to make $20,000 gold look NORMAL. And the timing is the real signal. This didn’t start during hype. It started AFTER the crash. When sentiment was broken and everyone was calling the top. That one detail changes everything. Because real money doesn’t chase headlines. It waits for stress. It waits for doubt. And then it builds quietly. So what does this mean? Someone with serious capital is STILL paying for extreme upside in gold. That’s not speculation. That’s preparation. I’ve studied markets for over a decade and called nearly every major market top, including the October BTC ATH. Follow and turn notifications on. I’ll post the warning BEFORE it hits the headlines.

有人刚刚开源了整个常春藤盟校的数学和科学课程。 斯坦福、麻省理工、哈佛……全部课程。集中在一个地方。完全免费。 Github：github.com/Developer-Y/ma…

You don't need the $100 Claude plan. Instead, you just need these 9 (free) fixes: 1. Shrink Your Files. Open a Google Doc. Paste the text from your PDF. Download as .md file format & upload that instead. A 15-page PDF = 45,000 tokens. The same text as the .md file = 2,000 tokens. 2. Plan First. Build Last. Open Claude Chat. Prompt: "Help me plan a [financial model]. Ask me questions first." Then paste the final plan into Cowork and prompt, "Build exactly this." You just saved 1/10th of the cost. 3. Let Claude Ask You. Stop writing long prompts. Prompt this instead: "I want to [TASK] to [SUCCESS CRITERIA]. Ask me questions using AskUserQuestion before you start." Then just select answers. Clicking costs almost zero tokens. Your 500-word prompt costs 500. Full guide here: ruben.substack.com/p/how-to-stop-… But here's what's really draining your credits: 4. Edit Your Messages, Don't Resend. Click "Edit" on your last message to Claude. Fix the mistake there. Hit Save. Claude regenerates without stacking a new message on top. Every "no wait, I meant..." doubles your cost. 5. Summarize Every 15 Messages. After having an entire chat with Claude. Prompt: "Summarize this entire conversation into a brief." Copy it & paste it as the 1st message in a new chat. You just compressed 105,000 tokens into 500. 6. Switch Models Before You Start. Click the model dropdown. If the task takes Claude under 30 seconds to answer, pick Haiku or Sonnet. Only pick Opus for deep, multi-step work. This saves you 3–5x per message. 7. Use Projects, Not Uploads. Go to Projects. Create one. Upload files there once. Every new chat inside that Project reads it without re-tokenizing. Stop uploading the same contract to 5 separate chats. 8. Turn Off Extra Features. Go to the tools panel. Turn off Web Search. Turn off connectors. Turn off Extended Thinking. Only turn on what this specific task needs. Idle features burn credits silently. 9. Batch 3 Tasks Into 1 Message. Instead of sending "Summarize this" then "List the key points" then "Write a headline" - send all three in one message. Three messages = three full context reloads. One message = one reload. Access the free guides to copy my full system: ✦ 23-fixes breakdown: ruben.substack.com/p/how-to-stop-… ✦ The infographic cheat sheet: how-to-ai.guide Send it to the one who keeps hitting Claude's limits.