coldfloat (@coldfloatrp) header
coldfloat (@coldfloatrp) profile picture

ai systems. automation that actually runs. less noise, more output

New York
Joined September 2026
Photo by coldfloat (@coldfloatrp)Photo by coldfloat (@coldfloatrp)Photo by coldfloat (@coldfloatrp)Photo by coldfloat (@coldfloatrp)Photo by coldfloat (@coldfloatrp)Photo by coldfloat (@coldfloatrp)Photo by coldfloat (@coldfloatrp)Photo by coldfloat (@coldfloatrp)Photo by coldfloat (@coldfloatrp)Photo by coldfloat (@coldfloatrp)
Machine Gun Kelly pulled Kelly Kay out of Logan Paul's party and did her on the back of Logan's G-Wagon. Standing up. Years later, she says she still has the hemorrhoid to prove it. In the video below, she tells the full story. She calls it the craziest one she has. It was the 100th episode of Impulsive, Logan's podcast. She'd recorded voice bits for their soundboard, so they invited her. She didn't really know anyone there. The guest that day? MGK. Nobody had planned it. After the show it turned into a party at Logan's house. MGK pulled her outside. He had one text he always sent her when he wanted her to come over. She says it was their version of "Netflix and chill." That night there was no bed. Just the back of Logan Paul's G-Wagon, standing up. Then she walked straight back into the party. A couple of days later the hemorrhoid showed up. She says she still carries the cream with her. "I love you, but you f*cked up my a**hole for life." Why did she go along with it? "I liked it because he liked it. I'll kind of do anything for my man."
21
2,191
"I'm a f*cking rock star and I want five b*tches around me at all times." That's what Machine Gun Kelly texted Kelly Kay when he ended things with her. In the video below, she tells the story for the first time. She says she'd barely even told her best friends. She says they dated for years before he blew up. Back then, he was "a famous failing rapper." By the time of this interview he was a rock star dating Megan Fox. Kelly says Megan is from a town over from where she grew up. One night after the Super Bowl she met him at LIV. He told her he was going to write a song about that night. She says he did. It's a Jack Harlow remix on his reels. But she'd get jealous. At one of his shows, surrounded by groupies, she drank whiskey and started crying. His answer: "I just have to provide for the party." His text said she was "wife material." That was the problem. Would she ever date a celebrity again? "No. There's just too much."
56
65
14
2,237
2,205,453
JAKE PAUL flew SKY BRI to Puerto Rico the day after his breakup with Julia Rose - and told her the whole thing was for clout, according to Sky Bri. 💀 She says she already knew him. Then Julia was out, and he hit her up. In the video below, she tells the story herself. The very first night, he flew her to Puerto Rico. The very first conversation was about clout. He wanted her to post it. She says he wanted to make Julia mad. "I knew that's what it was," she says. "It was very mutual." She had never been to Puerto Rico. She said she'd come if she could bring her best friend. He was into both of them. Then he pitched it: "Let's just do like a three-way relationship." They got each other's names tattooed. Hers is still there. Guys she's with now tell her to get it lasered off. What Jake did with his tattoo is the part that says everything.
1
10
2,749
Jordan Belfort says he lost his yacht and his private plane in the same 10 days. He had about $250,000 in cash on the boat. His friend went back for the quaaludes instead. In the video below, from a 2010 interview, Belfort tells the story himself, three years before The Wolf of Wall Street came out. It was 1996. He says he was already "four or five ludes deep" at the airport, and spent $80,000 on a new wardrobe for his wife on the way. The yacht was the Nadine, named after his wife. "It was Coco Chanel's boat," he says. The captain didn't want to go: 100 miles of open sea in a storm. Belfort: "Let's go." A 50-foot rogue wave flipped the boat. The front porthole gave way, and it started sinking. Six hours later the Italian Navy sent a helicopter. To make room for it to land, they pushed Belfort's own helicopter off the deck. "We can only take what we can carry," the captain said. Belfort had "like a quarter million in cash on the boat," plus artwork and jewelry. He sent his friend Rob back down. Not for the cash. For the drugs. Rob came back from the flooded, electrified cabin with a bag of 100 quaaludes and third-degree burns on his feet. Then they got to the airport. No private jet. It had crashed on takeoff, he says, after a seagull flew into the engine. The pilots survived. "You think I would have taken this as a sign that I should calm down... But I didn't, because I was addicted to drugs." In 1999 he pleaded guilty to securities fraud. He was ordered to pay $110 million back to the investors he defrauded. A yacht and a plane in 10 days. What would it have taken for you to stop?
1
9
605
Tom Dwan put $1.55 million into one pot with a single pair of queens. It became the biggest pot in televised poker history: $3.1 million. In the video below, he explains why he was sure Wesley didn't have aces. The game was Hustler Casino Live's $1,000,000 buy-in game in 2023. Wesley bought in for $3,000,000, more than anyone at the table. Wesley had ace-king. He bet $125,000 on the flop, $350,000 on the turn, and his last $786,000 on the river. Dwan called every time. Dwan says he picked up two of the biggest live reads ever on TV in that one 8-hour session. He didn't spend them on small pots. He filed them "in the back of my brain" and waited. "If he could have read my mind, he should have just run out of the casino and never bet another chip." Wesley tweeted after: "Sorry it didn't work out. 3.1M pot bluff. Biggest bluff in poker history." The same man who won $3.1M on camera has been accused by 3 players of owing them money. Peter Jetten says $226,000. Haralabos Voulgaris says $350,000. Dan "Jungleman" Cates says about $1,300,000. Dwan denies owing Jetten and Voulgaris. Poker genius, or the best player nobody wants to lend to?
4
14
727
i gave 6 grok agents my last $84 and one rule: no memes tonight, only real alts. by 8am it was $1,893 11pm. SOL, ETH, XRP, LINK, DOT, the boring stuff. 10-20x perps the floor: → ATLAS reads the chart → CHAIN watches whale flows on-chain → ECHO reads X → PULSE sits in telegram alpha chats → DEPTH watches the order book → GUARD trades nothing. he just approves or kills GUARD's checklist before any entry: funding neutral, OI not crowded, liquidation more than 6% away, spread under 0.02%, margin 60% max, stop -1.2% set before the order goes in 6,140 signals overnight. 10 got through the night: → 23:12 AVAX from ECHO, $50 in. +$40 → 00:58 LINK from PULSE. stopped, -$23 → 02:44 ETH from ATLAS. +$130 → 04:29 SUI from CHAIN. stopped, -$59 → 05:22 ADA from PULSE. +$281 → 07:07 LTC from ATLAS, $718 in. +$697 the weird part: each of the 5 agents found exactly 2 trades. ECHO 2, CHAIN 2, PULSE 2, DEPTH 2, ATLAS 2. nobody carried the night, the whole floor did and the wins kept getting bigger as the stack grew. size went from $50 on the first trade to $718 on the last. $40, $64, $92, $130, $184, $281, $402, $697 8 wins, 2 losses. both losers hit the -1.2% stop and that was it $84 → $258 → $572 → $794 → $1,893 that was all the money i had in that account. i went to bed expecting to wake up at zero bookmark this before you put your last $84 into a memecoin
Your backtest is lying to you and you don't even know it. This is the $200K quant stack — rebuilt with Claude in one evening. Full code. Copy it tonight or keep trading noise.
6
363
I gave my Grok Bot $1,870 and one task: make it 2x. It's 4:12am. It's at 8.64x. $1,870 → $16,157. +$14,287 in one night. 112 trades. 109 green. 3 red. It isn't one bot. It's four agents, each with one job: > SNIPER hits fresh pairs in the first 5 minutes, but only after LP is burned and mint is revoked > MOMENTUM waits for a volume breakout and rides it with a trailing stop > DIP HUNTER buys −30% wicks and sells the bounce > WHALE TAIL copies smart wallets the second they move Each one ran 28 trades. SNIPER, MOMENTUM and DIP HUNTER took one loss each. WHALE TAIL took none. The goal was 2x. It crossed that in the first hour, then 3x, 4x, 5x. Nobody told it to stop, so it didn't. The one red I saw live: SNIPER bought $MICHI, it dumped 6% in 3 seconds and the stop cut it at −$43. The other three didn't even blink. Here's what the last few seconds look like: -> DIP HUNTER buys $FWOG, dev didn't sell -> WHALE TAIL sells $PNUT, +$353 -> SNIPER sells $SPX, +90%, +$392 -> DIP HUNTER trails $FWOG at +24% -> DIP HUNTER sells $FWOG, +$627 Every trade is small. Every exit is planned. 97% of them land. It's SIM mode on live prices for now. I'm asleep. The four of them aren't. The whole build is in the post below ↓
5
231
Claude Code found 3 leaks in my trading bot. 48 hours later, $600 became $1,790... everyone's using it to find spyware. i pointed it at my bot instead 3 weeks on paper: 1,214 trades, 61% green, and the balance still went $1,000 → $868 one prompt, the code + 21 days of logs. it found 3 leaks: → take-profit was counted from the wallet i was copying, not my fill. i got in 3-9s late, so every "+30%" exit was really +4% → slippage at 0.25 since my first test, with my own comment next to it: "# temp, fix later" → 9 honeypots it bought and could never sell. $214 stuck 3 fixes. i didn't fully trust it yet, so i restarted on a fresh $600 paper wallet 48 hours later: $1,790. 84 trades, 58% green, one -17% dip on the way spyware is a one-time find. a leak in your bot costs you on every single trade check yours tonight: → is take-profit counted from your fill or from the signal? → search your config for "temp" and "todo" → can it sell what it buys? simulate the sell first save this and send it to anyone running a bot. prompt below ↓
1
1
4
183
the prompt, run it with claude code inside your bot's folder: "My trading bot wins most trades but the balance goes down. Read the code and the logs folder. Find every place where money leaks between signal and fill: entry delay, how take-profit and stop-loss are calculated, slippage and fees, tokens it bought but can't sell. For each leak give file:line, how much it cost in the logs, and the smallest fix. Don't change anything until I say yes." the last line matters. let it find, you decide what to change paper wallet, not financial advice
1
113
holy sh*t. i gave claude, grok, gemini and chatgpt $1,000 each and the same prompt. 30 days later claude had $38,420 and i'd bet on the wrong f**king one 4 paper wallets on live prices. every decision logged, all 4 on one chart grok led for 16 days straight. on day 11 it gave back 41% in one session and still stayed #1. i was already writing the "grok wins" post day 4 gemini was at $612. i had my cursor on the kill switch then day 17 claude took the crown and never gave it back chatgpt had the best win rate of all 4, 64%. it finished dead last with $6,840. 9.8s per decision vs 1.1s for claude. by the time it said "buy" the move was gone the first run died on day 2 because of that. chatgpt answered trades in paragraphs and my parser choked. after that all 4 had to reply in JSON only: buy / skip / sell, size, confidence. no prose $4,000 in, $81,930 out on paper picking buy or skip isn't a writing job. the one that explained the most lost who's your pick for round 2? reply and i'll run it prompt + rules in the reply
3
4
330
the setup, same for all 4: 1. one wallet per model, $1,000 each, paper fills on live prices 2. every 60s the agent gets state: balance, open positions, last 20 candles, volume, liquidity 3. it must answer JSON only, anything else = skip: {"action":"buy|skip|sell","size_pct":0-5,"confidence":0-1} 4. hard rules outside the model: max 5% per trade, stop -8%, no new buys after 3 losses in a row 5. confidence under 0.6 = skip, no matter what the system prompt: "You are a trading decision engine, not an assistant. You get market state. Reply with one JSON object and nothing else. If you are not sure, skip. Skipping is never wrong. Never explain." the "never explain" line is what fixed chatgpt. and it still came last
115
my last $40. told grok to pay for itself by noon tomorrow or it gets deleted. it came back with $3,861 the deal was simple: $3,500 in 24 hours or i unplug it. no extensions i gave it 8 wallets to copy. the weakest one made $388k in the last 30 days, the best one $741k first hour it was down $2.65. i already had my thumb on uninstall then it started hitting: → $NIHILISM, copied bigbabba, 12x, +$117. first rocket, balance jumps past $150 → at $1,338 it still needed $2,162 and the clock was running → $FTR, copied bystevenr, 17x, +$1,874. balance went from ~$1.3k to $3.2k on one trade → 52 losing trades along the way. most of them tiny, $BUBBLE -$3.02, $CLAUS -$5.05 → OmakaseOnly went big on $OTC. too big for my stack, so the bot just skipped it → ExactTallTakin alone made it $864 what it actually does: → all 8 wallets synced, copies a buy in 88–352ms → wallet sells, it sells. no hoping, no "one more candle" → bet bigger than the risk cap → skip, no debate → 143 trades closed, only 64% green → max drawdown -14.7%, never worse than that $40 → $164 → $353 → $1,150 → $3,861 i didn't click once. 105 copies, all on its own. by hand i'd be 2 minutes late on every single entry 64% win rate and still 96x. losses stay small, 5 rockets pay for everything. that's the whole trick paper wallet on live prices, so real fills will slip. the setup is in @0xboan's article below so yeah. it gets to live. for now bookmark this before you delete your own bot
4
127
f*cking insane. I hired an AI chief of staff for $0, gave it one rule: "$500 into $2,000 by morning or you're fired", went to sleep, and it had $2,275 on my desk by 8am a human chief of staff costs six figures a year. mine runs on Opus 5.5 inside Otto, on his own computer in the cloud, with my laptop shut i give the goal, he does the steps, i only make the calls the rule: "$2,000 by morning from $500. paper wallet only. 10% per trade, never lose more than 2% on one. no real money without my yes" the night: → 9pm broke the goal into 4 jobs and gave each a time, i wrote none of them → 11pm pulled 2,140 new Solana pairs and ran every one through 14 rug checks → 12:30am 1,912 dead. LP unlocked, mint on, dev holding 31%. one fail and it's out → 1am kept 25, each with a receipt for why → 1am to 7am 31 paper trades, 24 green. lost 3 in a row around 2am and didn't change a thing → 7am one page on my desk. live size left blank for me instead of guessing → 8am $500 → $2,275. I typed $200 for tonight honestly the blank live size is what I'd pay for. an AI that decides how much of your real money to risk is the one that empties your wallet and he does it again every night at 11, and he never forgets to start 2,140 pairs → 25 kept → 31 trades → $2,275 → 0 real dollars without my yes second 15, he leaves the live size blank and hands it to me someone's chief of staff is already trading the pairs you're still scrolling past paper fills, live will slip. I left the whole setup below ↓ free, steal it
1
4
122
I DON'T EVEN KNOW HOW TO REACT. 156 COPIED TRADES, 39 RED ONES, ONE FULL NIGHT OF SLEEP, AND MY $200 WAS $2,425 BEFORE THE DAMN ALARM EVEN RANG Put it on at 11:30pm. One target for the night: $2,000 by 7:30. Grok didn't pick one trader. It picked six, all from the top of the 30-day board, and copied them at the same time through copyfomo. Before every buy it says COPY or SKIP, and rugcheck kills the obvious rugs before Grok even looks. Every buy and every exit gets mirrored in about a third of a second. When a trader sizes in too big for my account, it just skips it. It said SKIP 70 times last night. The losses are real. 39 of them. At one point the account was down 7.9% from the peak, and I'm glad I was asleep for that part. But five trades did most of the night. $428 became $641 in one close. $1,625 became $2,203 in another. 117 wins, 75% hit rate. What I like most is how it split. The top trader made me $753, the sixth made $4. Nobody carried the whole night, and nobody sank it either. 3am the balance was $427. By the time I made coffee it was $2,425. Target hit, plus a bit. That's 12x on one night's sleep. I've traded crypto for years and every bad trade I ever made was at 2am. Turns out that's exactly the shift a bot should be working. Paper run on live prices, so real fills will slip. The full setup is in @0xNevsky's guide below, free. Most people won't save it, so you'll be ahead if you do.
i don't pick traders anymore. i don't pick coins. grok ranks the traders worth copying rugcheck kills the rugs before grok is even asked grok calls COPY or SKIP on every buy copyfomo fills it in ~1 sec, sells included full setup, every file paste below👇
4
122
I WOKE UP, OPENED MY GROK BOT AND JUST SAT THERE. THE $50 I LEFT IT AT NOON YESTERDAY WAS $2,975 ON PAPER. BETTER THAN ANY TRADE I'VE EVER MADE BY HAND 20 hours. 1,197 new memecoins killed, 43 bought. The buys are the least interesting number in it. Everyone screenshots the green. The part that made the money is the 96% it refused. Every new pair goes through 14 checks before a dollar moves: LP locked, mint and freeze off, dev under 5%, top 10 under 30%, no honeypot, tax under 3%, not a clone of a live ticker, and last, Grok reads X for the dev's history. Fail one and it's dead. No second look. I built it from @0xboan's article below in one evening. Paper mode, real prices, nothing touches a wallet. The five big closes, all at the TP it set before entry: - 03:18 $GLIMP, +112% move, +$36.88 - 06:51 $NUBBO, +131%, +$80.68 - 10:24 $FRAWG, +118%, +$186.09 - 13:57 $ZOOTI, +147%, +$510.11 - 17:31 $KELPO, +124%, +$1,144.99 Same size of move every time. The payout grew because the wallet under it grew. $KELPO's +124% paid 31x what $GLIMP's +112% did. 42 closes, 36 green. Avg win +4.9%, avg loss -0.9%. The reds stayed small because nothing gets in that can vanish in one block. Honest part: this is paper. Fills are simulated, and memecoins slip a lot harder live. And 14 checks still won't catch a dev who plays clean for six hours. Hour 20, 8 AM. $2,975 closed, one position still open. It said no 1,197 times, and the next rug is still most likely hiding in the 43 it said yes to. What would you add as check #15?
239
I TRIED TO OVERRIDE CLAUDE 4 TIMES THIS MONTH. I WAS WRONG ALL 4 TIMES, AND IT STILL TURNED MY $200 INTO $4,162 WITHOUT ONE TRADE FROM ME A month ago I put $200 in a fresh wallet and gave it to Claude. Day 1 I wrote a rules file and locked it, because I know myself. Day 3. Wallet at $188. I typed "stop, this is way too slow for memecoins". It kept going. Day 6 it was at $251. Day 9. A trader it copies aped a new coin and I wanted 25% of the wallet on it. Claude skipped it completely. LP wasn't locked. It rugged 40 minutes later. Day 14. It dropped the trader I'd copied by hand for a year. Three losses in a row and you're out, no matter how good your month was. I asked for him back. Nothing. He gave back 38% of his PnL the next week. Day 22. $2,040 on the balance. I wanted to pull it all and call it a win. The rules file doesn't care how big the number gets. Last time I asked. What it actually does: scores every trader on the fomo leaderboard on 30 day PnL, win rate and drawdown. Under 80, gone. 1,412 in, 9 kept. Every copy is 2% of the balance, auto take profit, trailing stop at 12%. 187 copied trades. 19 stopped out. 4 rugs skipped. 0 by hand. I've traded for five years. It didn't beat me on entries. It beat me on the 4 days I wanted to be clever. Paper wallet, simulated fills. Setup's below, free.
$200 → $4,287 in 30 days i didn't click a single buy. claude did picks traders, checks every coin, opens telegram, places the trade on copyfomo itself ~65 trades, 1 in 3 lost, still x21 full setup + every file 👇
2
121
gave Grok Bot $61 and went to sleep. woke up, saw $5,874, and my first thought was that the app was broken. I asked it for one thing: only take the boring trades. two hours in, one close had doubled the wallet. by the end, a +19% trade was paying 9 times what I started with. it took a few tries to get the desk right. six agents, and nothing gets bought until it passes the filters. thin liquidity, an unlocked LP, a dev holding too much supply, and the pair is skipped. whales get tracked, never copied. 273 trades, 231 green. the big closes, and the wallet under each one: > 02:22, FIZZO +$86, doubled an $81 wallet > 05:10, KABOO +$117 > 09:43, SNOOT +$214 > 13:26, WUMPY +$336 > 18:36, PEBL +$554, +19% on a $2,916 wallet that last line is the one I keep rereading. +19% is a boring trade. on $61 it would have paid about $12. by 18:36 the small stuff had been feeding the balance all day, so the same move paid $554. the feeding was 268 small closes, every single hour green, $4,505 between them after losses. the 5 big ones together made $1,309. and it barely gave anything back. profit in pieces, losers cut fast. 42 red trades all day, the worst was -$80. everyone is hunting the coin that does 50x. this run never touched one. its best trade by percent was the very first. $61 to $5,874 didn't need a moonshot. it needed a balance that kept growing under it. paper money, simulated fills. not financial advice.
My Grok Bot trading desk, fully open: → 6 bots, one shared machine → every charter, copy-paste ready → a read-only API key, so it physically can't trade without me → go/no-go numbers set before the backtest Bookmark it before you build yours.
2
74
THIS IS STARTING TO FREAK ME OUT I LET GROK COPY 6 WALLETS WHILE I SLEPT. $100 AT MIDNIGHT, $5,463.04 BY 8 AM. I gave Grok Bot one job before bed: copy the best memecoin wallets, and sell when they sell. With one change. When the wallet takes profit, the bot sells 70% with him. The other 30% stays in. I call it the rider. 4 hours in, 27 trades, the balance was $865.07. A stop on $WOOLY -$40.53, and the bot moved on. The money was in the riders. $BRRR: copied $101.76. The wallet sold at 2x, the bot sold 70% with him. The rider ran to 8.3x. +$263.57 $NARWHL: copied $549.45. 70% out at 2x with the wallet, 30% kept riding. 6h 42m into the night it hit 12.7x. +$2,148.36 Most copy bots exit when the wallet exits. Fine, until the wallet sells at 2x and the coin goes 12x without you. Selling 70% at 2x returns 1.4x your entry. The whole position is already in profit. The 30% left is free: if it dies, you still won. If it runs, it pays for the night. Final: 43 wins, 10 losses, 81% win rate, 2,744 tasks across six agents. $100 became $5,463.04. 54.6x. Two riders made $2,411.93. 45% of the profit came from the 30% the wallets left on the table. Paper money, simulated fills. Not financial advice. The full setup is below, free. Bookmark it before tonight.
1
96
I GAVE MY GROK BOT $100 AND ONE RULE: 2 LOSSES AND YOU'RE FIRED. 24 HOURS LATER: $3,662 The first trader it kicked out was the #1 wallet of the week. Firing him cost me $5.95. How it works: Grok Bot takes a trader from the weekly top, copies every buy at 0.25x his size, and sells the second he sells. Paper account, simulated fills. My only job was to watch. - 00:00, hires @mako_sol, #1 of the week - 05:15, two reds in a row, fired. Wallet at $185 - 05:15, hires @nori_ape, #3 - 12:36, two reds in a row, fired. Wallet at $330 - 12:36, hires @velvetbag, #2. At midnight the wallet reads $3,662.88 Then I replayed the same day with no rule. Same $100, same #1 trader, never fired. He kept trading after his two reds and drifted down. That wallet closed at $178.82. So the profit didn't come from picking the best trader. It came from leaving him on time. The #2, the last hire, carried it from $330 to $3,662. Two late sells paid $771 and $1,740. Two firings, $26.89 in losses that triggered them. That's the full price of the rule. 49 copies, 39 green. At 0.25x and out on his sell, the worst red was -$30.38. Hour 24. @velvetbag is still in the seat. One more red pair and he's gone too. Paper money, simulated fills. Not financial advice.
109
I STOPPED CLICKING. IT DIDN'T. I GAVE OPUS 5.5 A PAPER ACCOUNT AND A 16-LEG SCANNER THAT READS WALLETS, NEWS, ON-CHAIN AND SENTIMENT IN ONE PASS. PAPER RUN: $200 INTO $2,946 IN 10 DAYS. A human trader reads one thing, then the next, then sizes the trade. By the time you're done, the move is gone. This reads everything first and only then decides. 8 sections. 340 pairs across crypto, gold and silver. Around 9 scans every second. Every read feeds one score. 21 at the start. 47 after news. 66 after liquidity. 91 before execution. Nothing opens below 72. Risk per trade is capped at 3%. Then it acts. Stop to breakeven at +1R. Take profit at +3R. In one 4-second window it closed SOL, moved a BTC stop and opened gold. $312 after 14 trades. $806 after 31. $1,702 after 47. $2,946 after 58. 69% win rate. I lose a trade every time I check my phone. Claude doesn't have a phone. Paper account, simulated fills. Not financial advice. The full setup is below and free. Almost nobody has it saved. Save it before the next red week.
Your backtest is lying to you and you don't even know it. This is the $200K quant stack — rebuilt with Claude in one evening. Full code. Copy it tonight or keep trading noise.
1
109