That girl who opened AI "for five minutes" and is now building entire projects with it. @CELE8EX https://nitter.cf/t.co/cMuW6ZlEua
Joined October 2026
- Tweets231
- Following81
- Followers813
- Likes1K
Qweexee retweeted
Take an 18-year-old’s drugs away and this is what you get: he snapped and turned his fists on his own mother
→ then, when officers stormed in to pull him off her, he st*bbed three of them.
He sat there smug the whole time, smirking like it was all one big joke at everyone else’s expense. Then came the number — bail set at about $3.25 million . The grin dropped dead off his face.
THE BULLDOG REFUSED THE HAIRCUT AND GAVE ONE TO THE GROOMER INSTEAD
→ A bulldog on a grooming table shoves the groomer so hard he flies into the kennels
→ A second groomer steps in and the dog jumps up and hugs her
→ Next shot, she is face down on the table with the dog standing on her
→ The clippers are running through her hair, not his fur
207K likes and 16M views on the source
All of this is AI, and the physics are what sell it: the heavy body weight of the bulldog, the groomer's body flying into the cages, the paws pressing on her head and the clippers cutting through hair in close up.
Comic chaos with real bodies is one of the hardest things for AI video to hold together, and it stays consistent in the clip.
You can try Seedance 2.5 in @Picsart and build your own pet chaos scene
Which one of us would the bulldog groom next
Qweexee retweeted
Matt Wolfe built a full game from zero and showed every step in 28 minutes - ONLY 28 MINUTES!
Play the finished game: the-librarian-game.vercel.ap…
He says he's new to coding. The game is a reverse Vampire Survivors: you're a librarian, kids throw books off the shelves, you put them back before the chaos meter hits 100%.
Set aside 28 minutes. That's enough to watch it and start your own.
What you need (all free except Claude Code):
→ ChatGPT free plan to turn your idea into a design doc
→ VS Code, free, with Claude Code (comes with Claude Pro, $20/mo)
→ Leonardo for the sprites, 150 free tokens a day
→ Suno for the music, 50 free credits a day (about 10 songs)
→ free GitHub and Vercel accounts to put it online
In the reply: the full code of this game and every prompt he used, in order. Everything is ready. Watch the video, then go make something of your own ↓
Qweexee retweeted
Hey @grok, what sentence would you give this teenager if you were the judge?
A 14-year-old girl was shot in the back of the head by two boys she considered friends. They left her in a ditch, believing she would never be found alive.
But she survived.
Her name was Deserae Turner. And what she said to one of her attackers in court a year later made him break down in tears.
On February 16, 2017, 16-year-old Colter Peterson and his friend Jayzon Decker lured Deserae to a dry canal in Utah.
The plan was to kill her.
Peterson shot her in the back of the head at close range. Then the boys took her belongings and left.
Deserae lay there for nearly eight hours before two women found her.
She survived, but the bullet remained lodged in her skull. The damage left her with permanent disabilities. She struggled to walk, see, eat and even think clearly.
On February 1, 2018, Deserae came face to face with Peterson at his sentencing hearing.
She was sitting in a wheelchair. He was standing before the judge in handcuffs.
As she spoke, Peterson began to tremble and cry.
Then she looked directly at him and said:
“You shouldn’t have messed with me, because I am tougher than a bullet.”
But she wasn’t finished.
“Good luck in prison, and remember that because of you, my life is a prison.”
Then came the words that made the courtroom even heavier:
“Welcome to hell. I have been here for a year now.”
Peterson was sentenced to 15 years to life in prison. His accomplice, Jayzon Decker, also received 15 years to life for helping plan the attack.
Two boys tried to take her life. But when she finally faced them in court, she was the one who had the last words.
THE DELIVERY GUY WAS A GOLDEN RETRIEVER AND HE ATE PART OF MY ORDER
→ A dog in a red cap and a delivery jacket opens the door with a McDonald's box
→ He licks his lips, begs with both paws, then hops on a yellow scooter
→ Rides off down a rainy street with the bag strapped to his back
→ The box is open: two burgers with bites missing and a pile of fries
180K likes and 26M views on the source
Every frame is AI, and the part that sells it is how real the small things feel: the wet pavement, the cap sliding as he turns his head, the paws gripping the handlebars, the crumbs on his face.
Pets doing human jobs is the easiest idea in the world and the hardest to make look this natural, and this one holds for 29 seconds.
You can try Seedance 2.5 in @Picsart and make your own
Which job should the dog take next
THE HOST COULDN'T KEEP A STRAIGHT FACE AND THE WHOLE TALK SHOW IS AI
→ A platinum blonde guest in a black dress leans into a story on a late night set
→ Her face changes with every sentence: raised brows, eye rolls, hands moving
→ The host looks down at his cue cards, then covers his mouth to hide a laugh
→ He leans toward her at the end while she points at herself
6559 likes and 3.6M views on the source
None of this happened in a studio. The skin, the lip movement, the laugh behind the hand and the wide shot of the set with the moon backdrop all hold together for the full 15 seconds, and that is the hardest thing for AI video to fake: two people reacting to each other in real time. This is the kind of scene you can try with Seedance 2.5 in @Picsart
Would you have noticed it was AI without being told
SHE SKIED OFF A JUMP, THROUGH A HELICOPTER AND OUT THE OTHER SIDE
→ A skier in a white and pink suit speeds toward a huge snow jump in the Alps
→ A helicopter hovers right above the lip with its door wide open
→ She launches straight into the cabin and flies past the seats
→ Exits through the opposite door and keeps flying over the mountains
[likes] likes and [views] views on the source
All of this is AI, and the part that gets me is how the physics hold: the snow spray, the rotor blur, her poles and the tight cabin interior all stay consistent in one continuous take. A stunt like this would need a pilot, a ski team and a permit, and here it's one prompt Seedance 2.5 in @Picsart
Would you have believed this was real for the first five seconds
8
34
1,360
SHE JUMPED A MOTORBIKE INTO A FLYING CARGO PLANE AND CAUGHT A HANDLE MID AIR
→ Girl rides a dirt bike down an abandoned runway
→ Hits a ramp right under a huge low flying plane
→ Leaves the bike and grabs a handle hanging from the belly
→ Ends up dangling above the runway, laughing
1.6M views and 5,737 likes on the source
Every frame of this is AI, and that's the point: the motion blur on the runway, the wind in her hair, the jet engines overhead and the physics of the jump all hold together for the full 10 seconds. A stunt like this would need a stunt team, a plane and a permit, and here it's one prompt (Seedance 2.5 in Picsart)
Would you watch a full AI action movie built like this
SHE SPRINTED UP A CURVED DAM WALL, REACHED THE TOP AND SLID ALL THE WAY BACK
→ A full sprint straight up the curved concrete wall
→ A crowd lines the rim as she reaches the top
→ She slides back down in a trail of spray
→ She lands at the bottom laughing into the camera
Everything holds together in one continuous shot: the wet streaks on the concrete, the crowd leaning over the railing with their phones and the cameraman waiting at the bottom. Even a real person would be hard to tell apart from this
You can try Seedance 2.5 in @Picsart
Does anyone do it better than Picsart?
Qweexee retweeted
F*ck YES... Anthropic just dropped the most INSANE cheat code for Claude Code:
Fable 5.2 brain → Sonnet 5.5 builds → Opus 5.5 checks → Haiku 5.5 swarms
1 min + /advisor command = Fable-level decisions without Fable-level bills
first understand how it works:
• Sonnet 5.5 runs the session and writes the code. Opus 5.5 checks every diff. A swarm of Haiku 5.5 agents searches the repo, finds call sites and pulls the docs in parallel
• Fable 5.2 never writes a single line. It sits on call as the advisor and steps in at the moments that decide the outcome: before a plan locks, when an error repeats, before "done"
• you pay Fable prices only for the decisions, the cheaper models do everything else
and the numbers are from Anthropic's own advisor tests: Sonnet with an advisor came out 11.9% cheaper per task AND scored higher than Sonnet alone. Haiku with an advisor cost 85% less than running Sonnet solo
here's how to set it up:
1 → run claude update. Fable 5.2 needs Claude Code v2.1.321+
2 → paste the prompt below into Claude Code. It reads Anthropic's official advisor docs first
3 → it sets the team at user level, so it works in every project: Sonnet main, Fable advisor, Haiku swarm for subagents, Opus as the checker
4 → it backs up your settings and shows every change as a diff before it touches anything
5 → the part nobody tells you: Claude decides when to call the advisor on its own. There's no setting that forces it. One line in CLAUDE.md telling it WHEN to call Fable is what makes the whole setup actually fire
full prompt ↓
"Read code.claude.com/docs/en/advi… first. Then set up my Claude Code at user level:
1. Main model Sonnet 5.5. Set advisorModel to fable in ~/.claude/settings.json.
2. Create read-only Haiku 5.5 subagents in ~/.claude/agents: search (maps the repo), grep (finds call sites), docs (pulls the API reference).
3. Create an Opus 5.5 subagent called checker that reviews every diff and tries to break it, without editing files.
4. Add one rule to ~/.claude/CLAUDE.md: consult the advisor before locking a plan, after the same error twice, and before calling a task done.
Back up my current settings first and show every change as a diff. No edits until I say go."
on some plans Fable needs a one-time OK: run /model fable once, then /advisor fable
Fable thinks. Sonnet builds. Opus checks. Haiku swarms.
official docs: code.claude.com/docs/en/advi…
Comment "DN" and send me this post in DMs, I'll give you access to setup
SHE PLAYED FOOTBALL ALONE IN AN EMPTY CONCRETE HALL AND THE CAMERA COULD BARELY KEEP UP
→ A kid in white watches from the far end of the hall
→ She flicks the ball with her heel and spins around it
→ The shot cuts from right behind her to a wide view of the whole hall
→ One last kick sends the ball out of the frame
1.1M views. 3,695 likes. 418 comments
The ball bounces off the concrete the way a real ball does, the reflections on the floor follow her every step and the building stays the same across every cut. Even a real person would be hard to tell apart from this
You can try Seedance 2.5 in @Picsart
Does anyone do it better than Picsart?
THIS VOLCANO POPCORN IS NOT REAL
→ A man in a helicopter pours a bucket of corn straight into an active volcano
→ The kernels hit the lava and pop into fluffy popcorn in mid-air
→ The lava glow, the smoke and the helicopter frame stay consistent in every shot
Every detail holds: the glowing magma, the steam rising from the crater, the metal floor of the cabin, the way the kernels scatter. Nothing melts or warps between frames, and that is what makes it hard to believe it is generated.
You can try Kling 3.0 in @Picsart
Would you have guessed it was AI without being told?
SHE PUT HER HEELS ON THE HOST'S DESK AND HE KEPT READING THE CARD
→ A late-night set with the city glowing behind it
→ The guest sits down and sizes up the desk
→ Both heels land next to his microphone
→ He keeps reading, then breaks into laughter
None of this exists. The set, the skyline, the dress, two people talking in sync and the burned-in captions are all generated, and the host's deadpan reaction is what sells it. A reaction shot that stays in character while the scene changes around him is something video models used to fumble
You can try Seedance 2.5 in @Picsart
Would you have noticed it was AI without being told?
8
56
24,331
Qweexee retweeted
THIS INFLUENCER IS NOT REAL
→ A woman talks straight to the camera
→ Her lips match every word she says
→ The room behind her never shifts
3M views. 750 likes. 550 comments
Every detail holds: the skin, the hair, the light on her face, the hand gestures. Nothing melts or drifts between frames, and that is what makes it hard to believe it is generated.
You can try Kling 3.0 in @Picsart
Would you have guessed it was AI without being told?
A TALK SHOW GUEST WALKED IN BAREFOOT AND BROKE THE HOST
62k likes · 857 comments · 14 seconds
A tight close-up of a blonde guest in an oversized graphic jersey, fidgeting with the hem as she talks. Cut to the wide: she's barefoot on the couch beside the host's desk, a city skyline glowing behind them. She says something, he leans in, and both crack up.
Why it works:
→ Real studio camera logic. A locked-off close-up on the guest, then a fixed wide of the couch and desk. Nothing drifts like a handheld phone shot, so it reads as broadcast.
→ Classic cut order. Close-up first, wide second. That's how talk shows cut a moment.
→ Flat, frontal light. Studio lighting is meant to be even, with the skyline a touch darker behind. There's no dramatic shadow where an artifact could hide.
→ The laugh has a sequence. She talks, he reacts, she breaks and covers her face. Two people timing off each other is much harder to fake than one.
→ One exchange, one couch, fourteen seconds.
Takeaway: copy the camera grammar of the format you're recreating. People recognize a talk show by how it cuts before they ever look at a face.
Single-moment shots like this are where Seedance 2.5 shines, and you can run it right inside @Picsart
SHE HIT A GLOWING GOLF BALL INTO THE NIGHT SEA
→ A red ball glows in the sand by a lifeguard tower
→ One full swing and it launches into the dark
→ The ball flips from green to blue as it flies
→ She walks off with the club like nothing happened
The sand catches the red glow, the shadows stay locked to her body and the ball keeps changing color mid-air, all in a single phone-style shot. Light that moves and reflects on a surface used to be the fastest way to spot AI video, and this one holds up the whole time
You can try Seedance 2.5 in @Picsart
Would you have noticed it was AI without being told?
3
17
2,737
Qweexee retweeted
Weekly Perp DEX Recap #2 | Nov 1-8, 2026
Hyperliquid did $62.6B in 7D volume. Lighter, Variational and Aster combined did $44.3B 👀
Bigger news though. On Nov 3 the @CFTC cleared single-stock perps on registered US venues.
What caught my eye:
> @HyperliquidX: Nasdaq-listed Hyperliquid Strategies filed to become a CFTC-registered FCM. Only 180K $HYPE went to the team on Nov 6, not the 9.92M on trackers.
> @tradexyz: ICE, the NYSE parent, is taking a minority stake and will exclusively operate trade[XYZ]’s regulated US markets, pending approval. $650B+ volume in about a year.
> @Lighter_xyz: #2 by 7D volume with $19.5B. October hit $72.9B, $19.1B of it via Robinhood Chain. DCM + DCO filing still planned for November.
> @variational_io: #3 with $13.6B, public mainnet is close. My all-time average cost is $7 per point, and OTC reportedly cleared $37.5-44. That’s ~5.4-6.3x from my cost.
> @Aster_DEX: #4 with $11.15B. Fees bought back 2.6M $ASTER, and the same amount got burned from the team allocation.
> @arcus_xyz: Multi-asset collateral is live for everyone, stock tokens and wETH now back perps. My cost after Week 5 is $0.71 per point, still farming.
> @entropyIO: Holds ~51% of the $134M Anthropic pre-IPO perp OI, per @FalconXGlobal.
> @OndoPerps: The CFTC call is basically what Ondo lobbied for in August. Ondo Perps is still non-US only.
> @QFEX: OI $183M, +33% in 30D.
> @TxFlow_L1: OI +78% in 30D, Season 0 retro out with 10.76M points.
> @risextrade: OI up to $85M and still climbing. Fully onchain orderbook, spot is next. My cost is $0.91 per point and I keep stacking.
> @nadoHQ: PRIME, the 0% fee OTC venue, still targeted for Q4.
> @extendedapp: $XTEN tokenomics are out. 30% to points holders, fully unlocked at TGE. TGE date TBA.
MY TAKEAWAY
The US went from someday to a timeline:
> CFTC opened the door for stock perps
> ICE picked Hyperliquid rails via trade[XYZ]
> Hyperliquid Strategies and Lighter are lining up for licenses
Price doesn’t care yet. $HYPE was ~$90 in last week’s recap, now ~$77. Top traders across major perp DEXs were ~94% short on Nov 8.
Anthropic reportedly lists the week of Nov 16. That’s the first real test for pre-IPO perps. If it opens near where these books price it, that’s huge validation for Entropy.
My take: bullish on where this is heading, cautious on how fast approvals land.
None of this is guaranteed, and the IPO date isn’t confirmed.
Who benefits most if the US opens up?
More numbers, less noise.
Weekly Perp DEX Recap #1 | Oct 25 – Nov 1, 2026
12 projects. 7 days. A lot happened across the perp DEX ecosystem.
Here's what caught my attention this week:
> @risextrade: $3.6–3.8B in 30D volume, ~$21M TVL. Spot orderbook testing with ~10ms latency. Ignite Season 1 distributes 200K points weekly.
> @arcus_xyz: ~$63M TVL, with stocks accounting for ~56% of perp volume. Mobile early access launched in APAC, multi-asset collateral rolling out. @NYSE also spotlighted Arcus, highlighting its mission to bring global access to 24/7 trading, tokenized stocks and lower fees.
> @entropyIO: Crossed $100M OI and recorded $1.57B October volume. Building a major onchain pre-IPO market.
> @HyperliquidX: Still the leader with ~$222B 30D volume and $7.45B OI. Strong HIP-3 activity, HYPE trading around $90–91.
> @variational_io: ~$450B cumulative volume, $1.8B dual-sided OI and 580+ markets. $VAR TGE still targeted for Q4, with 32% allocated to points holders.
> @TxFlow_L1: ~$4.7B 30D volume and $47M OI. Season 1 distributes 1.25M trader points weekly.
> @Lighter_xyz: Preparing CFTC DCM + DCO filings for November, aiming to become a regulated onchain perp exchange. 100% of revenue goes to buybacks.
> @QFEX: Godel Terminal launched with 230+ stocks available 24/7. Loss Refund Program continues.
> @nadoHQ: Introduced PRIME, an onchain OTC-style venue for large trades with 0% fees. Launch targeted for Q4.
> @OndoPerps: Joined the DTCC Tokenization Service ecosystem. OpenAI-linked Private Markets subscriptions expected around Nov 2.
> @Aster_DEX : Expanded USDC-quoted BTC, ETH and SOL perps on Arc. OI remains around $1.4B.
> @extendedapp : Completed migration to Arc. Spot trading is live, while $XTEN tokenomics and whitepaper are expected in early November. 30% of supply allocated to points.
MY TAKEAWAY
Three narratives are getting stronger:
- RWA & pre-IPO: Arcus, Entropy, Ondo, QFEX and Nado.
- Regulated perps: Lighter is making a serious move toward the US market.
- Product expansion: RISEx, Variational, TxFlow and Extended keep building beyond basic perp trading.
Hyperliquid still leads the category, but competition is getting much more interesting.
Which project has the biggest upside from here?
More numbers, less noise. 🫡
Qweexee retweeted
This Polymarket trader has made +$337,337 in just 2 months, but he lost -$806,507 in the last 24 hours
He has made 56 trades with a Win Rate of 50%, and almost all of them are on football
Just a day ago his PnL was around $1.14M
His biggest wins so far:
→ Will BV Borussia 09 Dortmund win on 2026-11-04? | Yes at 65.3¢ | $1,886,019 → $2,888,582 | +$1,002,563 (+53.16%)
→ Will Chelsea FC win on 2026-09-18? | No at 64.9¢ | $1,762,013 → $2,713,484 | +$951,471 (+54%)
→ Will Brentford FC win on 2026-09-18? | Yes at 38¢ | $404,752 → $1,066,516 | +$661,764 (+163.5%)
Two of them came on the same day: he was against Chelsea and with Brentford on September 18, and both positions won
Football is his main sport, mostly the Premier League, La Liga, Serie A and European cups
Tennis showed up only a couple of times: he won on Andreeva vs Potapova at the US Open (+$33,913) and lost -$495,590 on Michelsen against Tiafoe
His positions are huge, from $300K up to almost $2M, and most of his entries are between 35¢ and 70¢
That’s why the swings are so big
His three biggest wins alone added up to about +$2.6M, but his total PnL is only +$337K, because his biggest losses are just as large
Celta Vigo cost him -$957,401, Como -$754,911, and Man United -$797,881
Most of his loss over the last 24 hours comes from that Man United position, Yes at 48.8¢
His biggest active trade right now:
→ Will FC Barcelona win on 2026-11-08? | No at 49.1¢ | $290,616
He’s risking almost $300K on Barcelona not winning, so let’s see if he can start recovering
What do you think?
Let me know in the comments ↓
This Polymarket trader just placed his biggest tennis bet so far, and it’s the biggest entry on the underdog in the Rolex Paris Masters final
He has been on Polymarket for only 2 weeks, made 52 trades and sits at -$33,633 overall, though he added +$726 in the last 24 hours
His biggest wins so far:
→ Spread: Cavaliers (-3.5) | Timberwolves at 50¢ | $10,000 → $20,000 | +$10,000 (+100%)
→ Trail Blazers vs. Lakers: O/U 233.5 | Over at 51¢ | $10,200 → $20,000 | +$9,800 (+96.08%)
→ Timberwolves vs. Cavaliers | Cavaliers at 53¢ | $10,600 → $20,000 | +$9,400 (+88.68%)
All three are NBA games, and most of his trading is focused on basketball, mainly spreads and totals
He also has a few positions on CS2, college football and football
He usually puts about $5,000 into a position at around 50¢, so his best win is about +$10,000 and his worst loss is about the same
Tennis has never been his thing
He had only two tennis trades before, and both were losses: Sakkari vs Eala (-$708) and Fritz vs Alcaraz in Paris (-$2,800), where he backed Fritz at 28¢
His active trade right now:
→ Rolex Paris Masters: Ben Shelton vs Carlos Alcaraz | Ben Shelton at 23¢ | $5,175 → $22,500
He’s going against Alcaraz again, but this time at 23¢ instead of his usual 50¢ lines
Alcaraz has lost only one set in Paris, and that was against Fritz
In that match Taylor had two match points, but Carlos came back and won in three sets
After that he beat Ruud in straight sets in the semifinal
Shelton, on the other hand, played three three-setters in a row against Rinderknech, Paul and Van Assche
He dropped a set in each of them and still found a way to win
Personally, I think Alcaraz wins this one
Ben’s serve can take him a set, but Carlos looks fresher, and he has already shown he can come back from match point down
Whatever happens, good luck to our trader
What do you think?
Let me know in the comments ↓
Qweexee retweeted
the guy who created Claude Code barely writes code anymore.
Boris Cherny built it as a side project in 2024.
by late 2025 he went a whole month without opening an IDE.
Opus wrote around 200 PRs, every single line.
so what does he actually do all day?
he hands the model three lines:
→ a goal: "make p95 ci time <10 mins"
→ a way to measure it after every change
→ how hard to try: "don't stop till it's done"
then he asks for a chart of progress every day.
CPU, memory, CI times, frame rates, latency.
anything with a number, he lets Claude hill-climb.
the part everyone skips:
he holds Claude's production code
to a higher bar than human code.
tests, fuzzers, adversarial review, every day.
the freedom only works because the meter is strict.
i put his whole loop on one page,
with a brief you can paste tonight.
if you can't measure it, you're still the one writing it.
