WHAT THE F**K. A GROK DESK SAID NO TO 23 OF 34 COINS, THEN TURNED 1.20 SOL INTO 14.64 SOL.
The rules on screen
7 checks, one red light kills the coin
Stop at -20%
Take half at 2x
Trail 15% from the peak
It said no to 23 of 34 new coins. Three of them
0:01 JASON6, top 10 wallets hold 41%, the cap is 30%
0:10 HODL, freeze authority set
0:13 MEOW, creator sold
0:06 GTA6 passed 5 of 5 and the desk bought.
0:06 to 0:08, the balance goes from 1.30 to 7.83 SOL.
0:16 to 0:19, from 9.70 to 14.64 SOL.
The 11 buys: two exits at +340%, one at +180%, one stop at -0.08 SOL. The other 8 split 4 breakeven, 3 wins, 1 loss.
At +340% I would have sold everything. The desk sold half and let the rest trail.
Would you take half off at 2x like this desk, or let the whole position run?
The setup is in my article below.
Test on a demo account first. Memecoins can go to zero and you can lose everything you put in. DYOR. NFA.
3
5
192
I can't f**king believe one prompt builds a whole pump.fun desk.
A 2026 REPORT: ABOUT 98% OF PUMP.FUN TOKENS SHOWED SIGNS OF A RUG OR A DUMP. THIS PROMPT BUILDS A DESK THAT SAYS NO FIRST.
The 4 images are the whole prompt: 5 parts, pasted into Grok Bot in order.
Radar finds every coin that graduates off the curve.
Inspector runs 7 checks. One red light kills the coin.
Judge scores it 0 to 100 with plain math.
Sizer caps the stake, so a stop costs 1% of cash.
Guard owns every exit.
No AI is ever asked to buy. Code makes every call, you get one button and you tap it yourself. No seed phrase, no private key.
Real money needs 2 weeks of paper, 100 closed paper trades and 9 passing tests.
The full setup and the Solidus Labs source are in my article below.
Test on a demo account first. Memecoins can go to zero and you can lose everything you put in. DYOR. NFA.
The 7 checks: mint authority, freeze authority, creator wallet, top 10 holders, liquidity, pool age, linked wallets. Which one would you delete first? Tell me and I will show you what it catches.
4
7
321
Replying to @maxt3chno
rockstar has 2 billion. bro has 10% quota and unlimited confidence
Deyv retweeted
Wayne Rooney told Manchester United on a Wednesday that he wanted to leave, and on the Friday he signed a new five-year contract for about double the money. That was October 2010. Sixteen years later he says the money had nothing to do with it.
In the video below, on Stick to United a couple of months ago, he admits “do I regret doing it? Of course I do,” and says he had just watched Tevez and Ronaldo walk out and wanted to know “what are we doing to replace” them before he gave the club “the biggest five years” of his career. When he asked, he says, “I got told where to go basically.” The reason he signed so fast in the end, he says, is that “there weren’t me looking for more money or anything like that.”
Sir Alex Ferguson told it very differently at the time. On 19 October 2010, the day before a Champions League game, he sat down in front of the cameras and said David Gill had come to him with “bad news”, that Rooney’s agent had let them know he wouldn’t sign and “he wanted to leave,” and that as early as March Rooney had been talking about wanting “a life contract.” The club, he said, was “as bemused as anyone can be.”
The next day Rooney put out his own statement saying he hadn’t been given the assurances he wanted about the squad, and with Manchester City being talked about as his next club, a group of United fans turned up outside his house in the middle of the night. By the Friday he had signed until 2015, “in the absolute belief” that everyone at the club was committed to winning, and Ferguson said “Wayne has apologised to me and the players.” The Guardian reported the new deal at around £180,000 a week.
Ferguson’s side is the second half of the video, and there’s one line in it most people forget. Before anything went public, he says, the club was already ready to give Rooney a contract that “would be difficult to better elsewhere.” Rooney’s version is that Ferguson just went public out of nowhere, “bang,” when they had “been having talks for months.”
I think they were both right, and that’s exactly why it blew up. Rooney asked a fair question about the squad, Ferguson heard a player threatening to leave, and once the word “leave” was out there, the only quick way to make it go away was a bigger contract. It doesn’t really matter now whether he wanted the money, because the money is the only part anyone remembers.
On the same show, the host asks who United actually went out and bought after giving him that contract, and nobody in the studio can come up with a name.
United fans, was Rooney fighting for the club’s ambition in 2010, and City fans, did he ever really mean it or was your club just the threat that got him paid?
This is f*cking insane.Ex-OpenAI agents engineer: "The trade is rarely the mistake. The click is. Wrong button, wrong size, stale price."
One night on one dashboard: 1,184 clicks checked, 4 denied, 0 bad clicks.
Five agents ask. Grok Bot checks. Only then does the cursor move.
Stale price at 0:02. The quote was 0.6s old. Denied.
Wrong size at 0:07. $420 instead of $42. Denied.
Wrong button at 0:11. BUY clicked, the ask was SELL. Denied.
Then 0:15. A price alert pops up right over Sell. Denied.
Every one was fixed and checked again before it passed.
The agents do not get the last word. The rules do: 1% risk per trade, stop at -20%, take half at 2x. Paper never slips, a live account can.
Which would cost you more on a real account, the extra zero or the wrong button?
The full desk is in my article below.
Test on a demo account first. Memecoins can go to zero and you can lose everything you put in. DYOR. NFA.
1
7
248
THIS F**KING AMAZING
A RULE-BASED DESK SAID NO TO A COIN AT 06:36, THEN YES AT 06:38. $95.00 TO $2,415.00 IN ONE NIGHT. 25.4x
Most traders decide on a coin once. Then they defend that decision no matter what.
My bot doesn’t have an ego. It only has rules.
885 coins checked. The Inspector blocked 18. Five agents, one line on the header: "code decides, no AI buys."
06:36, $Socks: red light, top 10 wallets hold 61%. Not traded.
06:38, same coin: migrated 4m ago, LP burned. Bought at 3% size.
06:46: take-profit sold it for +$187.39, 3.8x.
The rule did not move. The data did.
Losses stay small because the stop is -20% on every trade. Two from the log: -$0.74 at 23:55, -$1.55 at 02:10.
Final tally: 52 trades, 34 wins, 18 losses, 65% win rate. Best trade 9.01x.
Which rule would you trust first on a real coin, the 61% holder gate or the -20% stop?
The setup is in my article below.
Test on a demo account first. Memecoins can go to zero and you can lose everything you put in. DYOR. NFA.
2
245
A GROK BOT TURNED $67 INTO $2,244.50 ON GTA 6 NIGHT. 33.5x
The command: "GTA 6 drops tonight. find GTA coins with real news and data behind them. trade them all night."
It skipped two coins
GTA6: mint authority still on
Rockstar: one wallet held 49.89%
What it traded
WASTED, $20.00 at 10:07 PM, +$6.92
Official GTA, $16.00, stopped at 10:52 PM, -$5.07
GTAVI, 30% sold at 3:55 AM for +$495.72
GTA, $82.62 at 4:41 AM, sold at 5:10 AM for +$629.31
Tally: 11 trades, 5 coins, 83% win rate. Best trade +$1,449.65. Worst trade -$5.07. Stop at -30% on every trade.
A simulation has no failed transactions, no bots buying ahead of you, no gap when you try to sell. A live memecoin gives you all three.
Take profit at +761.7% like the bot did, or hold the rest to release?
The setup is in my article below.
Every screen in this clip says "simulation night recap."
Test it on a demo account first. Memecoins can go to zero and you can lose everything you put in. DYOR. NFA.
1
3
332
A simulated night recap on market : $50 became $3,500.00. 70x.
The command was one line: "Grok, run the night desk ." Session 23:40 to 09:40.
The rules in the prompt
Buy only if 2+ whales enter and the debate says yes.
Max 10% per trade.
Stop at -8%.
One coin it passed on: debate bull 2, bear 3.
The path
$75.13 at 1.50x. $242.03 at 4.84x. $653.86 at 13.08x. $2,184.85 at 43.70x. Then $3,500.00.
Best trade +$412.50. Worst trade -$41.30. Win rate 79%.
Paper is always kinder than real. No failed transactions, no bots buying ahead of you, no gaps when you try to sell. That is why the guide below starts on paper and puts three gates before one real dollar.
Which rule breaks first on a real memecoin, the -8% stop or the 10% size?
Test on a demo account first. Memecoins can go to zero and you can lose everything you put in. DYOR. NFA.
159
WHAT THE F**. I JUST GOT GTA 6 RUNNING ON MY PC WITH MY OWN AI.
I gave it one command:
RUN GTA 6 ON MY PC
It didn’t download some shady “GTA 6 PC” file.
It figured out the actual solution:
Stream the game from my console using Remote Play.
First, it tested my network:
5 GHz Wi-Fi
30 ms ping
Too laggy.
So I rewrote the code and optimized the connection.
PC on 5 GHz
Console on Ethernet
9 ms ping
Final result:
1080p
60 FPS
GTA 6 running on my PC.
No crack. No fake PC port. Just Remote Play and my own AI development.
Still an experimental build, but the result is real.
Would you let an AI set all this up for you?
3
1,524
Fable 5.2 cut the token price 20%. One analysis still found the cost per task at max effort roughly doubled.
The setup in this animation is the first fix I would test. Four roles:
- Advisor: the strongest model, called before a plan locks, when an error repeats, and before "done"
- Builder: a mid model that writes the code
- Checker: tests in the browser
- Swarm: small helpers that read files and git log
Every session starts with fresh context. The state lives in progress.txt, a feature list and git commits. 20 of 60 features passing, one commit per feature.
Top-tier thinking on every step is a fucking expensive habit. Pay for it three times a run, not three thousand.
Full guide below. Would you run your agent this way?
2
207
I can't f**king believe a model made this.
Third-person game. Oil paint look. A forest full of falling autumn leaves, aurora over the snow, a lantern in the hand, a cabin with a fire inside.
The clip's caption says Claude Fable 5.2 generated all of it.
Not a pitch deck. Not a mockup. A character you can walk through every scene.
Stop using Claude as a snippet machine. Give it a project and a finish line.
The setup that keeps the bill sane is in the full guide below.
Would you still pay a studio for the first prototype?
2
151
Fable 5.2 runs a team of agents on one landing page, and the clip looks fucking unreal.
It doesn't show the bill.
On the Artificial Analysis index, medium effort scores 56 at $3.80 per task. Max scores 65 at $26.54.
That is 7x the price for 9 points.
Medium already beats Fable 5.1 at max (56 vs 49, $3.80 vs $12.99).
Claude Code starts Fable at medium. The API starts at high. Same model, different invoice.
Max is a ceiling test, not a default. One setting decides what you pay.
The 7-step setup and a 30-minute test are in the full guide below.
Would you pay 7x for 9 points?
1
4
188
Andrej Karpathy built a full app. Then one prompt made the entire thing pointless.
He is a co-founder of OpenAI and the person who coined "vibe coding."
His problem was simple. Restaurant menus have no pictures, and he often had no idea what 30 to 50% of the dishes were.
Watch from 4:54.
So he built Menu Gen:
- upload a photo of the menu
- the app reads every dish title
- an image generator makes a picture of each one
- the app rebuilds the menu and runs on Vercel
Then he saw another version of the same idea.
Take the same photo. Give it to Gemini. Ask it to overlay the dishes onto the menu.
It sent back the exact photo he took, with every dish rendered straight into the pixels. No app. No code in between.
Timestamp 6:04. His verdict on his own project: it was all spurious. That app shouldn't exist.
Most people use AI to build the old thing faster. Karpathy says that is the wrong frame.
The neural network now does the work the app used to do. The input is an image. The output is an image.
So stop asking how to speed up what already exists. Ask what was impossible before.
Timestamp 6:27. The whole conversation is free online. Almost nobody will watch it.
2
3
274








