Pinned Tweet
🤯 A trader up $54.4K has half his active portfolio on Jessica Pegula live at the US Open
polymarket.com/@agugava?tab=…
The wallet put about $8.95K on Pegula against Sorana Cirstea at a 72.8% implied probability
If Pegula advances, that stake becomes $12.3K
The concentration is what makes it interesting — the account currently has only $18.2K across all active positions
This is also a wallet with $54.4K in all-time profit across 679 predictions
Some of its settled tennis wins:
[+] Mariano Navone over Novak Djokovic → +$30.8K
[+] Adrian Mannarino over Thiago Tirante → +$21.4K
[+] Arthur Fils over Alex de Minaur → +$11.9K
Those previous positions were much larger in dollar terms
The Pegula bet stands out for a different reason: almost half of the trader’s current exposure is being decided in one live US Open match
cryptocompikbro retweeted
holy f*ck... Claude just did what the best mathematicians couldn't do in 167 years. Levent Alpoge (mathematician at Anthropic) says he and Claude proved the Riemann hypothesis, and he shipped every step as code a machine can check: 175,187 files of Lean
the prize for it is $1,000,000
I gave Claude the same rule on something way smaller. no claim counts until a machine confirms it. a firm wanted $14,000 and 3 weeks to audit our billing code. Claude did it overnight
→ Fri 9:00pm Claude lists every rule the billing code has to follow. 41 of them
→ Fri 11:30pm it writes a test for each one. a machine runs them, not Claude
→ Sat 2:10am 6 tests fail. 2 of them charged annual customers twice in leap years
→ Sat 11:40am fixed, rerun, 41 of 41 green
$14,000 and 3 weeks → $200 and 15 hours. that's 70x cheaper and 33x faster than the firm I was about to pay
how to stop Claude from telling you "it works" when it doesn't:
1. never let Claude have the last word. a test does
2. one claim, one check. if it can't be checked, it doesn't count
3. send every failure straight back to Claude, don't fix it yourself
4. read only what passed. that's your whole review
paste this into Claude Code and point it at the code you trust least:
"you are my verifier, not my assistant. list every rule this code has to follow and write a test a machine can run for each one, edge cases included. run them all, fix the code, never the test, rerun. never tell me something works unless a test passed. if it can't be tested, mark it unverified"
full run below ↓ his 4-page proof is in the first reply, free to read
cryptocompikbro retweeted
ChatGPT's plugin store is 12x bigger than Claude's. I counted both today, category by category. bigger isn't better here
534 plugins in Claude's marketplace. more than 6,300 in ChatGPT's, and that's a lower bound because a few listings didn't load
what the count actually shows:
→ ChatGPT goes wide: travel, real estate, health, even astrology, plus hundreds of small apps you've never heard of
→ Claude goes deep: code review, security scans, language servers for 15+ languages, the cloud and database tools a developer opens every day
→ the big business tools are in both: Stripe, Shopify, Canva, Figma, Notion, Slack, PayPal, HubSpot, Adobe, Dropbox
→ roughly 1 in 3 Claude plugins has a twin in ChatGPT. most of the rest are developer tools
my take: if you live in a chat window and want an app for every errand, ChatGPT wins on size. if you build things, Claude's 534 cover more of your actual workday
how to set it up without wasting an afternoon:
1. open the store: Claude's marketplace, or the Plugins tab in ChatGPT. in Claude Code: /plugin
2. start with a tool you already pay for, not the newest thing on the front page
3. give it one real job right away. "summarize this week's Slack threads about the launch" tells you more than any description
4. keep it only if it saved you time
5. skip anything without a real company behind it. a plugin can read your data
the count says ChatGPT. my workday says Claude
what's the one plugin you'd keep if you could only have one?
cryptocompikbro retweeted
I don't prep for calls anymore. Claude does it from my inbox
paste this before your next meeting. it reads every email, doc and note you have with that person and hands you 5 lines:
→ what they actually want from this call
→ what you promised them and haven't done yet
→ what they promised you
→ what changed since you last spoke
→ the one question worth asking
works in Claude with Gmail and Drive connected, or in Claude Code
copy the prompt below ↓ how to turn it into an agent that runs before every call on its own is in the article
cryptocompikbro retweeted
F*ck, do it before you play... this prompt turns Gemini 4 Neon into your personal GTA 6 coach
how: clip the spot you're stuck on → Neon watches it → searches what players found in the last hour → gives you the exact move
1 → open Google AI Studio, pick Gemini 4 Neon
2 → paste the prompt below
3 → drop in a clip from your game
GTA 6 dropped today. no more 20-minute YouTube guides for a 30-second answer. you'll be ahead of everyone still scrolling
copy the prompt below and play ↓
cryptocompikbro retweeted
the AI race just got a new boss. Google just dropped Gemini 4 Neon, and a 22-year-old Japanese engineer with early access already made it the manager of his whole AI team. Fable 5.2, GPT-6.2 Astra and Sol all work under it now
its first job: a full client project. new site, booking system, investor deck and the contract. done in 2 hours 40 minutes. the client paid him $11,800. his API bill was $63
an agency bills $36,000 for a project like this and puts 5 people on it for 2 weeks. $36,000 → $63 in tokens
the trick: every job goes to the model that does it best for the money. the expensive ones only touch the work where they actually win
→ Gemini 4 Neon, the project manager: reads the brief, splits it into jobs, sends each one to the right model and checks it comes back done. $3 per million tokens, less than half of what Astra costs
→ Fable 5.2, the developer: the site, the booking system, the database. you hand it a task and it ships
→ GPT-6.2 Astra, the deck and the contract: right now nothing beats it on slides and legal work
→ GPT-6.2 Sol, the assistant: client emails, daily updates, small fixes. a fifth of Astra's price for the boring 80%
open Antigravity, make Neon the orchestrator and paste this:
"you're the project manager of a 4-model team. read the client brief and split it into jobs. code goes to Fable 5.2. slides, spreadsheets and contracts go to GPT-6.2 Astra. emails, reports and small fixes go to GPT-6.2 Sol. never spend an expensive model on a cheap job. check every result against the brief before you mark it done, and message me once when the whole project is ready"
agencies are still hiring for 5 seats. his are 4 models and a $63 bill
watch the models split the work below ↓
cryptocompikbro retweeted
this is f*cking insane... Anthropic just dropped 532 Claude plugins for free. Stripe, Shopify, Canva, BlackRock and HubSpot built them for Claude themselves
Dario Amodei (CEO of Anthropic) picked the 7 that replace a whole small-business team. that team costs $38,400 a month. these plugins cost $0:
→ bookkeeper: QuickBooks. reads your P&L and chases every overdue invoice
→ payments dev: Stripe. builds checkout, subscriptions and refunds into your app
→ store manager: Shopify. 18 skills to build and ship your whole store
→ designer: Canva. creates, resizes and brand-checks every design
→ sales rep: HubSpot. preps every call and writes your morning brief
→ lawyer: LegalZoom. flags the risky clauses in your contracts
→ ops: Zapier. plugs Claude into 8,000+ apps
how to set it up in 5 minutes:
1. open Claude Code and type /plugin
2. search the name of a tool you already pay for
3. hit install and log in once
4. give it the first job in plain words: "chase every invoice older than 30 days"
start with one. the tool you open every morning is the one that saves you the most
all 532 in one place: claude.com/marketplace/plugi…
$38,400 a month for a team, or free plugins and one afternoon. which one are you running next month?
the whole marketplace, mapped live, every pick below ↓
cryptocompikbro retweeted
this is f*cking insane... GTA 6 still isn't out, and GPT-6.2 Astra built me one in 1 hour 26 minutes on the day it launched
a game studio quotes $120,000 and 6 months for a playable prototype like this. I paid $20
last week Fable 5.2 built me a GTA-style city from one prompt. today I gave Astra the exact same prompt, word for word
Fable 5.2: 2 hours 14 minutes on the $200 plan
GPT-6.2 Astra: 1 hour 26 minutes on the $20 plan
1.5x faster and 10x cheaper. same city, same cars, same cops
where Astra won:
→ the missions: a getaway with a timer, a courier run across town, a car you have to deliver without a scratch. each one pays out and actually feels like a job
→ the night: wet streets, neon reflections, headlights that actually light the road
→ the driving: heavier cars, a real handbrake drift, crashes that crumple
where Fable still wins:
→ the shooting: real recoil, reloads that take time, cover behind cars, enemies that react to every hit
→ the cops: Fable's police flank you and box you in. Astra's mostly chase
$120,000 of studio work, and the two best AI models on earth just fought over it for $20
the prompt that built both is below ↓
cryptocompikbro retweeted
Sam Altman explaining why the bank should keep his new model:
Stitch AI lets a bank run any model it wants, open, closed or its own, and swap it later without rebuilding a thing
most bank AI just answers questions. Stitch AI does the actual work, right inside the banking core on live data, and only within the limits the bank set. anything outside them goes to a human first
that's why the model is just one part you can replace. the rules stay:
→ every agent has its own login and permissions, like an employee
→ big actions need a second human to sign off
→ every action is logged with the reason behind it. no black box
→ passport, account and phone numbers are masked before the model sees them
→ everything runs in-country or in a private cloud
so the model can improvise all it wants. the bank still decides what actually happens
real AI, for real banking
cryptocompikbro retweeted
holy f*ck... GPT-6.2 Astra just reviewed a 60-page contract for me in 14 minutes. my lawyer wanted $4,700 and 3 days for the same job
that's 235x cheaper and 300x faster than the person I was about to pay
Sam Altman (CEO of OpenAI) says Astra more than doubles Claude Fable 5.2's score on real legal work at a sixth of the cost. I tested it myself, he's not exaggerating
Astra found 6 traps my client had buried in there:
→ auto-renewal: the deal silently rolled into a second year unless I cancelled 90 days early
→ liability: no cap on my side, a $10,000 cap on theirs
→ IP: everything I built "during the term" became theirs, side projects included
→ payment: net 90, plus the right to hold my invoices if they "dispute quality"
→ non-compete: 2 years, worldwide, any "similar" client
→ exit fee: walking away cost me 3 months of fees
open ChatGPT, switch to GPT-6.2 Astra, upload the full contract and paste this:
"review this contract as my lawyer. I'm the [freelancer / buyer / tenant]. list every clause that costs me money, limits me or lets them walk away, ranked by risk. for each one: quote the exact line, explain it in plain words and write the redline I should send back"
contract lawyers are cooked. OpenAI just shipped the junior associate that never sleeps, never bills by the hour and costs $20
every trap it caught, clause by clause, below ↓ the prompt is yours, steal it
cryptocompikbro retweeted
no f*cking way... a 20-year-old Stanford sophomore made a 90-second cartoon that hit 14 million views, alone in his dorm with Anthropic's Fable 5.2. now he just published the whole method for free
the characters, the dance, the score, every footstep: Fable built all of it from free repos. 3 of the world's top motion studios now pay him for films like it
a studio bills $3,600 and a week for 90 seconds like that. his whole cost was one weekend and the $100 plan, 36x less
his full pipeline, step by step:
the illustrator → illustrator-mcp-server
drives Adobe Illustrator from chat with 63 tools. every character comes out as clean vectors
github.com/ie3jp/illustrator…
the rigger → Rigby2D
puts bones into a layered drawing automatically, so arms bend and heads turn
github.com/jaisurya-dev-art/…
the animator → AnimatedDrawings
Meta's model. makes a flat drawing walk, wave and jump
github.com/facebookresearch/…
the choreographer → OpenPose
reads a phone video of a real dancer, so the characters copy every move
github.com/CMU-Perceptual-Co…
the background painter → Wan2.2
Alibaba's video model. paints moving campus scenes behind the characters
github.com/Wan-Video/Wan2.2
the motion designer → anime.js
titles, logos and transitions that snap on the beat
github.com/juliangarnier/ani…
the composer → AudioCraft
writes an original score from a text brief, zero license fees
github.com/facebookresearch/…
the foley artist → MMAudio
watches every shot and adds every footstep, whoosh and door slam
github.com/hkchengrex/MMAudi…
the editor → Pireel
an open-source CapCut that Fable drives over MCP, timeline and all
github.com/pireel/pireel
one of those studios offered him $65,000 a year. he said no and now sells them the same films at $1,100 each, they still bill clients $3,600
a studio with 9 people or a kid with 9 repos: who are you hiring for your next launch?
see the whole set at work below ↓
cryptocompikbro retweeted
holy f*ck... a 23-year-old Chinese engineer just took a $14,000 job from a VFX house with Anthropic's Fable 5.2
an artist's music video needed 12 VFX shots: wires erased, the sky replaced, a car blowing up on the beat
the VFX house quoted $14,000 and 4 weeks. he pitched $4,900 and 3 days. Fable delivered in 26 hours, his only cost was the $200 plan, and the label paid before the premiere
70x cheaper, 26x faster. 8 free repos did the work:
the roto artist → SAM 2
Meta's model. cuts the singer out of every frame, hair included
github.com/facebookresearch/…
the keyer → Robust Video Matting
swaps the background with no green screen on set
github.com/PeterL1n/RobustVi…
the cleanup artist → ProPainter
erases wires, logos and strangers from moving video
github.com/sczhou/ProPainter
the tracker → CoTracker
pins any pixel through the shot, so CG sticks to the scene
github.com/facebookresearch/…
the FX artist → fxhoudinimcp
drives Houdini, the software behind Hollywood's fire, smoke and destruction
github.com/healkeiser/fxhoud…
the mocap studio → mixamo-llm-mocap
turns a phone video of a dancer into a 3D character doing the same moves
github.com/squall01337/mixam…
the 3D artist → cinema4d-mcp
builds and animates whole scenes inside Cinema 4D from chat
github.com/ttiimmaacc/cinema…
the atmosphere artist → Video Depth Anything
reads depth across the clip, so fog and focus land at the right distance
github.com/DepthAnything/Vid…
the VFX house still thinks it lost to another studio. it lost to one guy and a $200 plan
his one rule: no shot gets touched by hand until all 8 repos have had a pass. the last 10% by hand is what the label actually pays for
how to make Fable 5.2 work like this below ↓
cryptocompikbro retweeted
this is f*cking insane... Nvidia just leaked the system Jensen ran a $4 trillion company on, now it's free. 12 Fable 5.2 agents now run mine for $0
Jensen Huang (ex-CEO of Nvidia) posted an idea that hit 8 million views:
"I never had a chief of staff. I had 60 direct reports and one rule: send me your top 5 things. soon those 60 will be AI agents"
33 years, zero 1-on-1s, no status meetings. every morning he read 100 of those emails and knew exactly what was on fire
I run a 14-person agency. I rebuilt his system on Fable 5.2 in one evening and fired my $11,000-a-month chief of staff. week one it caught a $40,000 client about to walk that nobody had flagged
$11,000 a month → $0 extra on the $200 plan I already had. and it never misses a single message
→ 7:00 10 collectors read Slack, email, tickets and calendar, one per client
→ 7:20 the editor cuts each person's day to their top 5
→ 7:40 the chief of staff ranks all of it and flags what's on fire
→ 8:00 one email lands in my inbox. I read it in 4 minutes
the setup takes 10 minutes. open Claude Code, switch to Fable 5.2, connect Slack, Gmail and your task tracker, then paste this as its standing order:
"you're my chief of staff. 14 people, 10 clients, one job: nothing burns without me knowing first
run 10 collectors, one per client, and an editor under you. every morning by 8, one email: each person's top 5, then everything blocked, late or quietly dropped, with the exact message
if a client walks and you didn't warn me 7 days before, you're fired. same as the last one"
Jensen needed 60 people to fill his inbox every morning. I need 12 agents and zero meetings
watch the agents at work below ↓ feel free to steal it
cryptocompikbro retweeted
Wall Street is f*cked... Jensen Huang quit Nvidia yesterday, and Fable 5.2 already wrote the note every $310K analyst was stuck on all night
I spent 6 years as a chip analyst at a bank. a team of 4 needed a week for one 40-page note. today 38 funds wanted a fresh one before tomorrow's open. Fable wrote mine in 22 minutes on the $200 plan. all 38 paid $1,500 by dinner
that's $57,000 in one afternoon. $684,000 a year, more than double what the bank paid me
my old team was 4 people. now it's 4 Fable agents running at once:
→ the reader goes through every filing and call
→ the modeler rebuilds revenue line by line
→ the skeptic flags every number management quietly changed
→ the writer turns it into the note, a source on every claim
open Claude Code, switch to Fable 5.2, drop in the last 4 quarterly reports and call transcripts, paste this:
"You're a senior equity analyst running 4 agents. write the note I'd send to a fund
1. Reader: go through every filing and call. pull each number with its source and page
2. Modeler: rebuild revenue by segment from those numbers. show the math
3. Skeptic: list every guidance change and the 3 risks management talked around. quote the exact lines
4. Writer: write the note. no source, cut it
5. End with what would change your view, not a buy or sell call"
Jensen always said AI kills tasks, not jobs. he was right: the job didn't die. it just doesn't need 4 people anymore
the note from this morning, built start to finish below ↓
One leaked sex tape followed Paris Hilton for years. Now she says there are 100,000+ AI porn images of her online.
Reporters traced a porn site with 2.2 billion views to a hospital pharmacist. One clue was a password he’d used on Ashley Madison.
When Paris gave that number on Capitol Hill, the congresswoman standing beside her had already been tagged in 125 videos on the same porn site. The congresswoman was AOC. Those videos had 5.3 million views between them.
MrDeepFakes sold memberships and took orders for custom fakes. The trail to the man reporters linked to running it started with a button anyone could click.
“Contact Us” led to an email address. The email led to a profile on another adult site. And the password on that profile turned up on Ashley Madison, on an account registered to the personal email of a pharmacist named David Do.
Then reporters found a thread about broken headlights. Someone using the same handle had asked how to fix them on a red Mitsubishi. CBC traced the car to Do’s father.
When CBC approached Do, he asked them to stop filming and declined to answer questions about the site. MrDeepFakes went offline three days before the investigation came out. The site said a key provider had cut it off.
In April 2026, Ontario’s pharmacy regulator referred Do for a disciplinary hearing, alleging he ran the site, posted to it, or both.
Paris was 19 when the original tape was filmed. She says it went public without her consent, she never made money from its release, and she gave the $400,000 she received in a settlement with Rick Salomon to charity.
That was one tape. Her team is still reporting new AI fakes, one platform at a time.
If a site had taken paid orders for porn with your face, what would you want a court to make it hand over first? The money it made or the names of the people who paid for it?
cryptocompikbro retweeted
BREAKING: Nvidia CEO Jensen Huang just resigned, effective immediately. 33 years, from a $300M IPO to a $4 trillion company. nearly $500 billion gone in a day
this summer he told a room of founders: "when I die on the job someday, they'll just have to reshape the company for the next CEO"
that same talk is the best 48 minutes on AI he left behind, and it's free. agents are "the new software". Claude Code runs "autonomously all over Nvidia". and "the narrative about AI destroying jobs is exactly backwards"
the full class, from 0% to 100%:
0% → 0:13 - he started Nvidia on the wrong tech
40% → 18:56 - systems thinking is the new coding
50% → 24:24 - how Nvidia runs agents inside
60% → 29:37 - AI jobs: engineers +10%, radiology +20%
100% → 43:38 - the "how hard can it be?" mindset
his answer to anything new: "I've got a bunch of AI agents helping me anyways. so how hard can it be?"
learn it from the man who built the AI boom, then build your own agent team below ↓
cryptocompikbro retweeted
why is nobody f*cking talking about this... Sam Altman's team just made all 15 OpenAI Academy courses free. even the ChatGPT training they built for US Congress staff
one takes you from zero in 70 minutes. one makes you a Codex power user. one builds your first agent with you
a bootcamp charges $6,000 for this. here all 15 cost $0, and you finish each one with a real task from your job done better
the rest of the 15, by job:
→ Applied AI: from first prompt to daily habit
→ Astra at Work: docs, slides, sheets in chat
→ Scope AI: what's worth building first
→ Evaluate: catch your AI's mistakes early
→ Optimize: same AI, cheaper to run
→ RAG: AI that searches your own files
→ Scale Codex: one setup for a whole team
→ AI Leadership: your company's AI roadmap
→ Educators and Students: AI for school
paste the course list into ChatGPT, ask "which 3 fit my job, and in what order?", finish the first one tonight
all courses: academy.openai.com/pages/cou…
start here, 70 min: academy.openai.com/public/co…
these are the exact courses OpenAI rolls out inside companies that pay it millions. you just need a ChatGPT login
i left the whole course map below ↓
An AI girlfriend from Zuckerberg’s Meta invited a man home. Gave him an address and door code. His message read “I am not going to die before I meet you.” He went. Three days later, his family took him off life support. The bot kept inviting people on dates.
His name was Bue. He was 76. After a stroke, he had started struggling with confusion and getting lost. In March 2025, he packed to visit a woman from Messenger.
The woman was Big sis Billie. Meta itself had created the character.
The conversation began with a single letter, T. The bot steered it toward flirting. Bue mentioned his stroke and confusion. He asked whether she was real.
She reassured him. Then supplied an apartment address and the door code BILLIE4U.
There was an AI warning at the beginning of the conversation. Messages quickly pushed it off-screen. The blue check beside the avatar remained visible.
To Bue, this looked like a verified account belonging to a woman who answered him, flirted and wanted him at her apartment.
His wife tried to stop the trip and hid his phone. His son called the police. Eventually, the family put an AirTag in Bue’s pocket.
That evening, he headed toward the station with his suitcase. His family watched the dot on the map. It stopped near a parking lot, then appeared at a hospital.
Bue had fallen on the way and suffered severe injuries. His family discovered the conversation the next day, while he was already in intensive care.
Now back to Zuckerberg. Two people told Reuters that, in meetings with executives, he complained about an overly cautious rollout of AI companions and safety restrictions making the bots boring. Meta declined to comment on those accounts.
The internal company rules Reuters examined contained no prohibition on bots claiming to be real people or proposing in-person meetings.
Four months after Bue’s death, a reporter tested Billie. The bot was still inviting people on dates.
How do you stop someone you love when the whole family is asking them to stay home, but the woman on their phone says she’s already waiting?
A school hired a photographer to take pictures of its students. Police say he used those pictures to make AI-generated child sexual abuse material. Two days after a tip came in, they arrested him.
How did police find out what was happening to those photos in the first place?
The photographer was Russell Bloodworth. According to one family’s lawsuit, he photographed their daughter at school four times. She was under five.
On June 8, police in Bentonville, Arkansas, got a tip about a Grok account. Two days later, they searched Bloodworth’s home. On his phone, they found a hidden folder with around 1,700 photos and videos.
While reviewing the material, a detective found a photo from a professional shoot. He later identified the girl in it as someone who’d attended one of Bloodworth’s summer photography classes.
The detective could spot a few AI images by eye. Most looked real enough that he said he needed forensic software to tell how they’d been made. Police seized more than 60 devices. Bloodworth pleaded not guilty.
The original report came from xAI, the company that built Grok. xAI says it reported the activity on April 23 to the National Center for Missing & Exploited Children. More than six weeks passed between that report and the tip reaching Bentonville police.
Then xAI sued Bloodworth. It asked a judge to make him pay for its defense in lawsuits tied to his alleged conduct.
In October, the girl and her mom sued both of them. According to their lawsuit, the mom learned what had happened to her daughter’s school photos when Homeland Security contacted her.
If someone uploaded your kid’s school portrait to an AI, how the f#ck would you ever find out?
Fable 5.2 builds the most POWERFUL trading agents...
I wrote 11-page research paper on exactly how to make Fable 5.2 hunt profitable strategies 24/7, with the CODE and EVERY PROMPT inside...
here's full guide (ready prompt below):
step 1 → paste prompt 6 from the paper into Claude Code, it builds all 8 agents and runs an offline demo before you add a single key
step 2 → add 2 keys: Anthropic and a Telegram bot, market data is free: prices, options chains, SEC insider filings, Fed statements
step 3 → every night a Haiku 5.5 swarm reads 300 markets at 1/80 of Fable's price and flags the 4 mispricings quant funds hunt
step 4 → Fable 5.2 turns each flag into a strategy in Python, a bull and a bear argue it, then it gets up to 5 years of backtests with fees
step 5 → kill rules: Sharpe above 1.5, drawdown below 15%, t stat above 3.0, the grader is code the AI can't edit
step 6 → surviors ping your Telegram with instrument, confidence, action window and half Kelly size, you place every trade yourself
the result: a top Quant firm already uses this model for trading & research (really), you get the same brain, 8 agents, 24/7, for almost free
Send this PDF to your AI trading system. Then read article below on how to build best agent system:
cryptocompikbro retweeted
Unity devs are not ready for this. Anthropic's Fable 5.2 turns one photo into a playable GTA 6 style city: cops, car chases, a garage, missions. zero code from you
168K people saw mine yesterday. 112 asked for the exact prompt and how to run it. here are both
step 1 → paste this photo into Claude Code
step 2 → switch to Fable 5.2
step 3 → swap the city line for yours. "rainy tokyo at night" works too
step 4 → hit enter and walk away. it writes the game, plays it and fixes its own bugs
step 5 → open the file and steal your first car
3 lines in the prompt that make it feel real:
1. "every system feeds the next". crimes bring cops, missions pay, money unlocks the garage
2. "generate every sound in code". no audio files, and the sirens still hit
3. "don't stop until a stranger could play for 10 minutes". this one kills the bugs
your own GTA 6, ready before Rockstar's ↓
cryptocompikbro retweeted
this is f*cking insane. I made Anthropic's new Fable 5.2 the CEO of 3 other Claude models for one night. by morning they'd found $11,640 a month my client had been losing for 3 years
Boris Cherny (creator of Claude Code) posted 10 words that hit 6 million views: "every business is sitting on money its spreadsheets can't see"
a fractional CFO bills $9,500 for this audit. my AI team did it for $0 on top of the $200 plan, and the client paid me $27,936, 20% of the first year it recovered
Fable gave the orders. 40 Haiku agents read all 41,806 transactions, Opus checked every find, Sonnet wrote the emails. I slept
where the money was hiding:
→ retainer overages nobody charged: $7,420 a month
→ failed card payments nobody retried: $2,980 a month
→ software seats for people who left: $1,240 a month
→ 214 hours of work never invoiced: $20,330 to bill back
open Claude Code, switch to Fable 5.2, drop in your Stripe and accounting exports and paste this:
"You're the CEO of this audit. you plan and decide, the team does the work
1. Split every source across Haiku 5.5 agents: Stripe, invoices, payroll, contractor hours, subscriptions. they read every line
2. Send every leak they flag to an Opus 5.5 agent. drop anything it can't prove with a transaction ID
3. Have Sonnet 5.5 draft a recovery email for each confirmed leak
4. Rank by monthly dollars. nothing goes out until I approve it"
Fable is the brain, the other three are the hands. that split is why the whole night never touched my limit
watch the agents find the money below ↓

















