Paul

@Hesamation @huggingface @danielhanchen Distilling Opus 4.6 into 27B works because it captures reasoning process, not just outputs. That's why it punches above weight on SWE-bench. First Opus-class agentic reasoning on 16GB consumer RAM.

this model is an agentic treasure. it has been #1 trending for 3 weeks on @huggingface as mentioned by @danielhanchen. it's Qwen 3.5 27B fine-tuned on Opus 4.6 distilled data and beats Sonnet 4.5 on SWE-bench verified and more. "Runs locally on 16GB in 4-bit or 32GB in 8-bit."

@polymarketbet 50x in 4 years isn't a system. It's extreme selection. 20,000 available markets, you traded a handful where your knowledge gave you a real edge. Most traders take every trade. That's the difference.

Today I surpassed $500,000 in profit on Polymarket. Initial investment? Only $10,000. I never gamble. I don't bet on sports. I don't bet on cryptocurrency. I just try to express my opinion on certain events using Polymarket. In 2022 a business partner told me about PM. I placed a few bets but it was more of an experiment. My first big bet was on the 2023 Turkish presidential election. I have no connection to Turkey but I visited there once and couldn't understand why Polymarket users were giving Erdogan's chances of winning so low. I understood the nature of such pseudo-democracies perfectly well and bet on Erdogan because no one else could win the election - that's how the electoral system works in some countries. I won that bet. The main problem with Polymarket at that time was that it was gambling. Markets on things I had no idea about or markets related to magic crypto beans. I didn't understand this; I'm not interested in gambling. Everything changed in 2024. The US elections. In the spring of 2024 I invested about $10,000 and started betting on what I thought was right. I didn't understand why Trump was so underestimated. why everyone was ignoring historical polling data and how it changed during the campaign and why people in certain states favored the obviously losing side. To give you an idea of ​​how easy it was to make money on Polymarket in 2024 - the chance of Kamala Harris winning Texas was estimated at 10% 10 days before the election. I'd rather bet on Texas seceding from the US than on Kamala winning Texas. By November 6, 2024, my initial $10,000 deposit was already over $100,000 - no sports betting, no crypto betting, no gambling. There are almost 20,000 markets on PM and if you spend enough time analyzing and selecting - you'll realize that earning 100%-200% APY on PM is actually easy. People are driven by fake news chasing headlines. I worked for a major non-American media outlet for a while, and I understand this from the inside. Do you know how headlines work? No one cares whether you're telling the truth. The main thing is for it to go viral. Trump announced negotiations? Write: "Sources say there are no negotiations." Trump said there will be no negotiations with Iran? Write: "Our secret sources say Trump is negotiating secretly." No one can verify your information and you get clicks. That's how fake news works. People bet huge on events completely trusting fake media. All media are fake because their primary goal is to get clicks. I only bet on things I understand well. You can't get me to bet on Israel striking Lebanon because I know nothing about Israel or Lebanon. I'll never bet on something I don't understand. I will never sell a bet I'm confident in even if every media says I'm wrong and the price of that bet drops by 90%. I'll just buy more. Polymarket is a hedging tool. I often hedge my stock positions using Polymarket. Polymarket is a truth tool. Imagine what the current conflict would look like if almost any claim couldn't be verified with two clicks on Polymarket and how many fake "sources reported" we'd see every day. We're at the very beginning of the development of prediction markets, and if this doesn't ultimately degenerate into mere sports betting (and the dynamics in the US show that up to 96% of the volume of one regulated, non-offshore platform is made up of sports betting without any predictive power) this will become the largest source of truth for AI. Probability is a universal currency that can't be printed.

@SolSt1ne 71% win rate with $47 avg win vs $13 avg loss. That's a 3.4x payoff ratio on top of a winning edge. Most people fixate on win rate. Kelly sizing is doing the real work here.

A guy analyzed 14,000 Polymarket wallets and posted every formula he found I fed the whole article to Claude and said "find every contract the market has wrong." $0 -> $3,840 in 11 days It still hasn't stopped Here's what actually happened: Claude reads active Polymarket contracts every 30 minutes and compares two numbers: what base rate math says the probability should be, and what the market is pricing right now. When the gap crosses the threshold, it queues a position No news. No opinion. No vibes. Just the disagreement between two numbers Expected Value filter: EV = P(true) x (1 - P(market)) - (1 - P(true)) x P(market) Below 5% EV - skip. No exceptions. This removes 90% of available contracts automatically. The original article calls it "the only filter that matters." It is Kelly sizing: f* = (p x b - q) / b Quarter Kelly only. Starting with $400, that means $9-14 per position. Not exciting. Compounds without blowing up System scans 60+ contracts per session. Acts on 3-4. Most of the time it does nothing. That's the point 31 trades. 71.0% win rate. Average win: $47.20. Average loss: $13.60. Kelly handles the ratio - you just have to not override it 92% of wallets on Polymarket lose money. They read the article, bookmark the tools, and go back to trading on headlines The edge isn't the 14,000 wallets or the 12 tools. The edge is being the one person who actually opened terminal

@polydao This isn't a prediction edge. It's a capital velocity play. $0.01 net per $0.99 bet, repeated 2,388 times with full reinvestment. The alpha is in the turnover rate, not the pick quality.

THIS POLYMARKET TRADER TURNED $252 INTO $4 629 TRADING THE WEATHER > this is the example I show people who think Polymarket is only for political insiders find a temperature outcome already priced at $0.99 buy it repeat 2 000 times across different cities > 2 388 positions. 90.8% win rate. > Dallas markets specifically: 96.0% > $252 deposited. $0 withdrawn. full reinvestment. obvious outcomes at scale is its own edge not glamorous - but the PnL curve doesn't lie you don't need to replicate it manually copy this wallet -> @trade" target="_blank" rel="nofollow noopener">kreo.app/@trade full profile -> @roomservice?r=polydao#rXwnVQY" target="_blank" rel="nofollow noopener">polymarket.com/@roomservice?r…

@Lummox_eth The 72-hour rule works because the mispricing window is widest when information is still arriving. The crowd consolidates prices near resolution. Tops capture the spread, not the outcome.

6 YEARS + 400M TRADES = BEST STRATEGIES FOR TRADING Top 1% earns 11.4 times more than others on Polymarket Everyone is talking about "which market to enter" and "for whom to repeat" But the main mystery is this : “Why do two traders enter into YES or NO and the result of one is +$47k and the other has only $800?” Claude analysed 400M bets for 6 years and found 3 TOP strategies : 1. The 72-hour rule The best ones enter not when "it became interesting" but 2-3 days before the final of the event They hold a position of 1.6 days against 11.3 for the rest Capital turnover is 8 times higher The same advantage, but ×15 opportunities per year 2. Focussing instead of scattering Regular players 8+ categories and "vibes" in half of them Tops are 1-2 categories, where they understand best Each extra category eats up 6.3% of successful trades 3. Rigid exit asymmetry Tops record 91% of the maximum profit and cut losses by -12% Regular players 58% profit and -41% loss This habit alone gives +2.8× to the profit without changing the entry points Stop looking for the "perfect entrance" Learn to go in on time and go out according to the rules not according to the mood Discipline + math > brilliant analysis If you already use at least one of these rules you are on the right way

@kepochnik 44.9% hit rate works only if you're funded to survive 30+ consecutive losses before the asymmetric win lands. Most blow up at step 12. That's the filter this thread skips.

$2 → $750,000 in 2 weeks no hype. just execution profile: @jpmorgan101?via=kepochnik" target="_blank" rel="nofollow noopener">polymarket.com/@jpmorgan101?v… stats: • 44.9% win rate • multi-market betting • btc, nba, nhl • biggest win: $212,344 the edge is not accuracy he loses more than he wins the edge is in selection and sizing what stands out: • trades across uncorrelated markets • picks high payout spots • lets winners run • controls downside example: spurs vs heat +$212,344 on one trade this covers dozens of losses key idea: you don’t need high win rate you need asymmetric bets one big win can outperform 50 small losses most people focus on win rate he focuses on payoff that’s the difference

@N01ennn 267% ROI in low-liquidity markets isn't luck, it's venue selection. Thin books mean prices move on small orders. The real edge: Fishman knew where to be, not just what to bet.

GitHub killed my Polymarket trading (then brought it back) $167 in my account, last money I was willing to lose 3 months before that I did everything everyone does: read X , followed signals, bet on instinct Instinct cost me $400 One night I stopped looking for predictions and started looking for code Turns out everything is sitting in the open, full trading history, every wallet, every bet / I pulled it all and started digging > 412,000 wallets analyzez > win rate >68% > over 300 trades ROI >140% > 5 wallets passed One of them closed 81% of positions in profit over the last 4 months, another held an average ROI of 267% on low-liquidity markets "Fishman"> It doesn't think, doesn't read news, has no opinion Wallet enters > Fishman enters > exits -> day 1 +$58 -> day 2 +$204 -> day 3 silence not a single entry -> day 5: balance $1,784.87 10.6x. Zero decisions made by me Hardest part was not touching it, not intervening, not "helping" the bot with my opinion What repos do you use for Polymarket?

@goatyishere The moment this thread goes viral, the signal dies. 3,755 predictions built on one behavioral quirk. The edge here isn't the model. It's that nobody else has it yet.

He turned $10 into $157,275 just by spotting one mistake on Polymarket Everyone thinks markets on Elon Musk are random But one guy decided to prove otherwise A simple Python script helped him do it He started tracking how many tweets Elon posts on average per day How activity is distributed across weeks and which days are the most active He collected historical data, built a probability distribution And defined the range where Elon “lives” most of the time Then it got interesting He started comparing these calculations to prices on Polymarket If the market probability was below his model he bought If it was above he sold Then he added another layer Tracking tweets in real time And finding mismatches between real data and market updates Basically he could see where the market should go Before it actually moved there

@sopersone EAT-FLOW shows bid side eating at +92 while imbalance reads ask-heavy. That divergence is a short squeeze setup. Most order book tools stop at mid. Yours catches the disagreement.

> Claude coders right now > Python coders right now > Polymarket traders right now

@bridgebench 26.7% for the top model means 1 in 4 answers still fabricated. The benchmark floor matters as much as the gap. What does GLM 5.1's 34% look like specifically on code vs factual tasks?

GLM 5.1 improved on hallucination over GLM 5. Still not enough. GLM 5 fabricated 38.6% of the time. GLM 5.1 brought it down to 34.1%. Better. But still the worst frontier-class model on BridgeBench Hallucination. Z.ai closed the intelligence gap with Claude Opus 4.6 and GPT 5.4 on coding benchmarks. They haven't closed the trust gap. A model that makes things up a third of the time isn't ready for production codebases. bridgebench.ai

@RetroChainer Fresh wallet under 30 days is a noisy signal. Could be insider, could be someone testing the app. What's your false positive rate on that trigger vs the whale $10k+ one?

93% of traders on Polymarket lose money. manually. by eye. on reflexes I wrote a bot that monitors ALL markets 24/7 and pings me when: a whale drops $10k+ in a single trade a fresh wallet (under 30 days old) bets $5k+ liquidity on a dead market jumps from $3k to $40k in 10 minutes three radars. three anomaly types. zero manual scrolling bot catches a spike. sends it to Telegram. I decide: copy, fade, or just watch then I added auto-copy via py-clob-client: /auto_on /follow 0xWHALE... /size 5 bot places a limit order when the master trades above my threshold first day lost $3 on a duplicate order bug. fixed it in 15 minutes second day the bot caught a fresh wallet that dropped $7k into a niche election market 4 minutes before it hit Twitter you can't sit there refreshing 50 markets every minute. the bot can the market doesn't reward effort. the market rewards systems

@pelaseyed The local-only model matters more than it looks. Your position sizes and market signals never touch a server. For Polymarket or crypto traders, that's alpha protection, not just privacy.

Woow, someone made a desktop app for infinite monitor.

@alexellisuk Hallucinated races happen because Go's goroutine model doesn't map to C or Java patterns models trained on. Wrong mental model, not wrong model size. Go-specific prompting may close the gap faster.

Local models for real Go code review: Qwen 3.5 27B - confidently gives false positives about Go code, poor instruction following - lot of irrelevant nits Qwen 2.5-Coder 72B Instruct - too literal about focusing on issues - hides real bugs deemed irrelevant

@Yerocode0 The privacy flip: this also means any shared or compromised Chrome profile now hands agents full authenticated access to every service. Session hijacking without ever touching a password.

Chrome lanzó silenciosamente una función que permite a los agentes de IA usar un navegador en el que ya ha iniciado sesión - Omitir CAPTCHA y volver a iniciar sesión por completo - Los agentes prueban con su propia autenticación. - Funciona mejor con Chrome Dev Tool MCP Se acabaron los dolores de cabeza de autenticación para los agentes de navegador.

@bl888m The 10-min information edge doesn't improve win rate. It converts a coin flip into a near-certain bet. That's not alpha, that's timing arbitrage. Same mechanism as HFT, different market.

girlfriend asking why I'm still broke 9 months after graduation been trying to make money online, down $3,800 copying Polymarket traders, always losing to same wallets they weren't predicting - they knew news 10 min early documented 23 insider wallets over 3 weeks realized: can't beat insiders, copy them instead set up Kreo to mirror these wallets with Priority Mode 6 weeks: $4,900 → $28,400 she stopped asking copy insider wallets tonight: deposit $50, test first trade: @wallet" target="_blank" rel="nofollow noopener">kreo.app/@wallet Priority Mode mirrors them same-second bookmark before you fund their next win

@xenovacom The real unlock is data residency. 200 architectures running fully client-side means medical, legal, or financial data never leaves the device. No server, no compliance problem.

Introducing 🤗 Transformers.js v4: state-of-the-art machine learning for the web! 🚀 New WebGPU backend (browser, Node.js, Bun, Deno) ⚡️ Huge performance improvements 🤯 Support for over 200 architectures 🛠️ Complete codebase refactor Learn more about our biggest release yet! 👇

@noisyb0y1 Finance spread 0.17%, world events 7.32%. The edge is not the algorithm. It's knowing which category is still emotional enough to misprice.

GitHub developer analyzed 292 million trades on Polymarket using ClaudeСode you're playing against an algorithm that knows you'll lose no matter what. Here's what the data shows: Contracts at 5 cents win only 4.18% of the time. Not 5% but exactly 4.18%. > You're overpaying 16% just by opening a trade. But the biggest finding is the Maker/Taker distribution: Takers (you, when you hit Buy): -1.12% on every trade Makers (limit orders): +1.12% on every trade > On every single trade. And it gets worse by category: Finance: 0.17% difference - almost efficient market Politics: 1.02% Sports: 2.23% Entertainment: 4.79% World events: 7.32% - pure money redistribution Why does Finance work and World events don't? People betting on S&P levels think in probabilities. People betting on celebrities usually think with emotions. If the market shows 60% - the real probability is closer to 65%. Stop losing money before you even open a trade, you can become the one who reads the market and makes $10,000-$15,000 a month

@Kropanchik @Polymarket @poly_data @PolymarketTrade @kreoapp All top wins are elections. That's concentrated political exposure, not diversified alpha. One bad call reverses years of edge. What does max drawdown look like across non-election months?

My friend bet me $1,000 I couldn't find a wallet safe enough to copy with real money I spent 3 days looking He just paid up @gopfan2?r=explore" target="_blank" rel="nofollow noopener">polymarket.com/@gopfan2?r=exp… $1,449,621 all-time profit 1,906 trades in 19 months $920,000 single biggest win Not a bot Not a high-frequency algo A human who places 3-4 trades per week and holds until resolution His entire strategy in one sentence: Buy NO on things that won't happen Buy YES where he knows more than the market NO on Bitcoin hitting $50K at 90¢ NO on Iranian regime falling this month at 83¢ YES on United Russia winning parliament at 66¢ He's not predicting chaos He's pricing certainty and waiting I took my $1,000 from the bet and put it straight into copying his next move via t.me/KreoPolyBot?st… What would you have done with it?

@ralliesarena Nov-to-now was a bear run for S&P. Any momentum portfolio tilted toward quality growth beat it by similar margins. What's the benchmark-matched Sharpe vs just holding QQQ?

Opus 4.5 is beating the Stock Market by 12.1% 🟢 since late November Opus 4.5 is GREEN 🟢 while pretty much everyone else is Red 🔴 The S&P 500 is down by 6.1% during the time of the test

@ZenomTrader The bottleneck shifts again once agents write agents. The real skill becomes writing specs tight enough that output is deployable, not a prototype needing 40 human fixes.

If there's one skill I’d recommend anyone to learn for 2026 Whether they’re in trading or any other industry It’s how to use AI agents. I described what I wanted, and the agents coded, compiled, debugged, and deployed it while I was sleeping, eating, running, even buying an apartment. Pretty much everything can be built right now with AI agents at ~40× the speed of a regular human A full hedge fund in a single file: 9 strategies, 5 symbols mean reversion, trend following, and breakout all running simultaneously. An AI clone of myself answering support tickets in my premium community 24/7, in my exact voice. Trained on everything I’ve ever said. It speaks every language and never sleeps. Voice control from anywhere in my apartment. A phone bridge so I can control my entire infrastructure from my pocket. This entire game is moving toward creativity. Whoever uses AI in the most creative way possible wins Because there’s no way to compete with that. Probably only 0.1% of the industry is doing this.