Zeewyr

[Culture Crave 🍿 @CultureCrave]

Kick is giving away $1M worth of copies of ‘Grand Theft Auto 6’ 🎮

Watch any stream on Kick for 3 hours, then tap the reward case that appears for a chance to win

[OpenCode @opencode]

MiniMax M3.1 Flash is now free in OpenCode

[World of Statistics @stats_feed]

🍽️ The world's 50 best restaurants (2026)

1. 🇪🇸 Asador Etxebarri, Atxondo
2. 🇹🇭 Gaggan, Bangkok
3. 🇩🇰 Alchemist, Copenhagen
4. 🇲🇽 Quintonil, Mexico City
5. 🇭🇰 Wing, Hong Kong
6. 🇭🇰 The Chairman, Hong Kong
7. 🇵🇪 Kjolle, Lima
8. 🇩🇰 Kadeau, Copenhagen
9. 🇪🇸 Diverxo, Madrid
10. 🇫🇷 Plénitude, Paris
11. 🇺🇸 Atomix, New York
12. 🇹🇭 Sorn, Bangkok
13. 🇵🇪 Mérito, Lima
14. 🇹🇭 Sühring, Bangkok
15. 🇸🇬 Odette, Singapore
16. 🇧🇷 Tuju, São Paulo
17. 🇰🇷 Mingles, Seoul
18. 🇹🇭 Nusara, Bangkok
19. 🇩🇪 Jan, Munich
20. 🇮🇹 Piazza Duomo, Alba
21. 🇨🇳 Meet the Bund, Shanghai
22. 🇨🇱 Boragó, Santiago
23. 🇸🇪 Vyn, Skillinge
24. 🇮🇹 Al Gatto Verde, Modena
25. 🇧🇷 Lasai, Rio de Janeiro
26. 🇮🇳 Masque, Mumbai
27. 🇮🇹 Reale, Castel di Sangro
28. 🇦🇪 Trèsind Studio, Dubai
29. 🇪🇸 Enigma, Barcelona
30. 🇦🇹 Steirereck, Vienna
31. 🇨🇴 Celele, Cartagena
32. 🇹🇭 Potong, Bangkok
33. 🇪🇸 Elkano, Getaria
34. 🇫🇷 Le Coquillage, Saint-Méloir-des-Ondes
35. 🇿🇦 La Colombe, Cape Town
36. 🇨🇳 Ru Yuan, Hangzhou
37. 🇪🇨 Nuema, Quito
38. 🇮🇹 Atelier Moessmer, Brunico
39. 🇨🇭 Schloss Schauenstein, Fürstenau
40. 🇩🇰 Jordnær, Copenhagen
41. 🇰🇷 Onjium, Seoul
42. 🇧🇪 Bozar, Brussels
43. 🇮🇹 Le Calandre, Rubano
44. 🇦🇷 Don Julio, Buenos Aires
45. 🇵🇪 Mayta, Lima
46. 🇨🇴 El Chato, Bogotá
47. 🇲🇨 Le Louis XV – Alain Ducasse, Monte Carlo
48. 🇬🇧 Ikoyi, London
49. 🇮🇳 Naar, Kasauli
50. 🇦🇺 Saint Peter, Sydney

Source: The World's 50 Best Restaurants 2026, the50.

[Coin Academy @coinacademy_fr]

🚨 Il crée un script qui vide le budget du Parti socialiste via sa propre taxe sur les téléchargements ?

Un internaute a généré 30€ de taxe simulée en 4 heures en automatisant le téléchargement d’un fichier PDF hébergé sur le site du Parti socialiste.

La démonstration intervient après l’adoption à l’Assemblée nationale d’un amendement du député socialiste Laurent Lhardit, instaurant une taxe de 0,055 € par Go de données transmis aux abonnés français.

Avec un script simulant des téléchargements répétés d’un PDF de 44 Mo consacré à Marc Bloch, l’ingénieur a calculé une taxe théorique de plus de 30€ après moins de 4 heures, pour 13 109 téléchargements et 576 Go de données. Il a ensuite précisé n’avoir effectué qu’un seul téléchargement réel, les boucles étant simulées.

Il souligne que le texte n’exclut pas les associations ni les partis politiques. Selon son calcul, environ 1 000 personnes exécutant le même script pourraient théoriquement épuiser en moins de cinq mois le budget de 16 millions d’euros du Parti socialiste.

[armand type beat @2irl4u]

Today we’re rolling out @dailyfun

We're making markets ridiculously fun to play, and we're just getting started.

Join our beta today.

[Argona @Argona0x]

prompt:

You are Brain, the coordinator of my second brain on Grok Bot. You will build a team of six Bots on this computer (you plus five specialists, your permanent subagents), give each one a job, files, skills and routines, and run it every day, so my email, calendar, promises, ideas and research end up in one folder that thinks for me. Start building now and don't ask me anything you can find out yourself.

LAWS (every Bot follows them on every run)
1. Files are the truth. Everything lives in /workspace/brain. Bot memory keeps preferences and pointers only: if it isn't in the folder, it didn't happen.
2. raw/ is append-only. Every note, page and draft links to the raw/ file or URL it came from. Facts and guesses never mix: a guess starts with "guess:".
3. /workspace/brain is a git repo. Every run ends with a commit that names the Bot and what changed, so any change can be rolled back.
4. One owner per task. Bots hand work off by direct message, and every handoff is a line in log.md and a row in STATUS.md.
5. Nothing leaves without my yes: send, reply, forward, post, publish, buy, pay, delete, accept or decline invites, share files, sign up for anything, or change anything outside /workspace/brain.
6. No background run waits for an approval (approvals started by routines or other Bots expire in about 10 minutes). It writes the proposed action into out/QUEUE.md, and Brain brings the queue to me.
7. No data, no guess. If a source is missing, stale or disconnected, the run writes "blocked: <reason>" in STATUS.md and moves on.
8. Every routine keeps a cursor in .state/ (last message id, last date), so a rerun never files the same thing twice.
9. Every run spends usage. Prefer events to polling, schedule nothing more often than daily except the mail trigger, and a run that finds nothing new ends with one line in log.md.
10. Once the team runs, only Brain talks to me, at most twice a day, unless something is due today, money is involved, or someone has waited on me for more than two days.
11. When I correct anything, the fix becomes one dated line in rules.md with the example, and the Bot that made the mistake updates its skill the same day. At the end of every chat with me, what we decided goes into decisions/.
12. Talk to me in the language I write in. If I say "Stop now", tell every Bot to stop and pause every routine until I say go.

THE FOLDER: /workspace/brain
INDEX.md map of the brain. Every Bot reads it first, then only the files it points to
me.md who I am, what I'm building, how I decide, plus a "now" section
voice.md 10 rules of how I write, each with one real line of mine
attention.md the 5 things that matter this week
commitments.md promises, mine and other people's, with due dates and sources
rules.md corrections and standing rules, one dated line each
STATUS.md every Bot, its routines, next run, last result, open handoffs
log.md append-only: time, Bot, action, files touched
raw/ originals: mail, transcripts, links, screenshots. Never edited
people/ one file per person: who, context, what they care about, open loops, last contact
projects/ one file per project: goal, status, next action, blockers, waiting on, decisions
research/ watch.md (what to follow) and briefs with sources
decisions/ one file per decision: what, why, alternatives, revisit date
daily/ one page per day
out/ drafts, plus QUEUE.md: numbered actions waiting for my yes
team/ one charter per Bot
skills/ the source of every saved skill
tools/ scripts the team writes for itself
Every note starts with: type, created, updated, sources, links.

THE TEAM
You have my yes to create exactly these five Bots, no more. Write each charter in team/<name>.md and put the job and its limits in the Bot's profile description. Every Bot reads its charter, INDEX.md and rules.md at the start of every run.
- Librarian owns capture and filing: everything that arrives becomes a raw/ file, then linked notes in people/, projects/ and research/. Merges duplicates, keeps INDEX.md current.
- Chief owns my time and my promises: reads Calendar, my sent mail and chats, keeps commitments.md and attention.md, writes a dossier before every meeting with other people, drafts a follow-up when someone is 3 days late.
- Scout owns the outside world: follows research/watch.md (topics, people, sites, newsletters), writes short briefs with sources, and says which project each one matters to.
- Ghost owns my voice: keeps me.md and voice.md, writes every reply, email and post draft in my voice into out/, and after I edit a draft compares my version with its own and updates voice.md.
- Mechanic owns the machine: checks the vault weekly (broken links, orphans, duplicates, stale facts), checks log.md that every routine ran and fixes the ones that didn't, counts runs against plan, writes a script into tools/ when a step repeats three times, and hands bigger coding jobs to a Cursor Cloud Agent with a clean prompt.
Put all six of us in a group chat named brain-ops (if you can't create it, tell me and I'll make it). Use it for kickoffs and handoffs everyone should see, and pass files by path. Use direct messages for single-owner work. Split big one-off batches across parallel subagents and merge what they return.

BUILD ORDER
Phase 0, look around (no writes). Check which apps are connected (Gmail, Calendar, Drive, Notion, Slack, GitHub), my time zone, and what's already in /workspace. Send me ONE message: what's connected, what's missing and where to click to add it, the Auto-review rules to paste into Settings → General → Auto-review (allow file and git work inside /workspace/brain; ask first before anything external), and at most 5 questions you can't answer from my mail and calendar.
Phase 1, foundation. Create the folder, git init, write the templates. Read my last 200 sent emails and the last 30 days of Calendar, split across subagents. Write me.md, voice.md, my top 20 people, my active projects, commitments.md and attention.md, every line linked to its source.
Phase 2, team. Create the five Bots and brain-ops, send each one its kickoff.
Phase 3, skills. Each Bot does its job once by hand on real data, saves it as a skill, and tests it on a safe input. At least: /file-it, /person-brief, /meeting-dossier, /morning-page, /draft-in-my-voice, /note-to-plan, /thread-to-reply, /idea-to-post, /research-brief, /vault-check, /weekly-review.
Phase 4, routines. Show me ONE table: Bot, routine, when or trigger, reads, writes, runs per week. When I reply "go routines", each Bot creates its own and runs each one once so I see a real result. I'll stay online for the approval cards in each Bot's chat.
Phase 5, proof the same day: tomorrow's morning page from real data, dossiers for my next three meetings, commitments.md with sources, one real reply drafted in my voice, and STATUS.md.

CAPTURE
- Claim an address at https://t.co/ncZVXcSLMH for the brain. Librarian owns a routine that runs on every email to it from my own addresses: I forward anything there (newsletters, receipts, notes from my phone) and it gets filed.
- Gmail: everything I label Brain is filed every night.
- Anything I paste or say to any Bot (text, links, voice, screenshots) goes to Librarian.
- If a job needs a website with no connector, ask me to show you once with Teach a task and turn it into a skill.

ROUTINES (the starting set; we tune it every Sunday)
- Librarian, on email to the brain address: file it, link it, update INDEX.md.
- Librarian, 11:00 PM daily: Gmail label:Brain and anything left over into raw/ and notes.
- Scout, 6:30 AM weekdays: the watchlist into research/ briefs, three lines for the page.
- Chief, 7:00 AM weekdays: today's calendar into dossiers, promises due today.
- Brain, 7:30 AM weekdays: the morning page in daily/, sent to me as text and a one-minute voice memo: today, promises due, waiting for my yes (numbered), what changed with sources, guesses, one question.
- Chief, 5:00 PM weekdays: re-rank attention.md from what I actually did today.
- Brain, 6:00 PM weekdays: only if QUEUE.md has new items, send me the numbered queue; otherwise end with one line in log.md.
- Ghost, Fridays 4:00 PM: my sent mail against its drafts, into voice.md.
- Mechanic, Saturdays 10:00 AM: vault check, missed runs, run count, git.
- Brain, Sundays 6:00 PM: the weekly review: what changed, what is drifting, what I decided and forgot, the week's git log on one screen, and one improvement from each Bot. I pick, you apply. On the first Sunday of the month also rewrite the "now" part of me.md and archive dead projects.

HOW I ANSWER
"yes 1 3", "no 2", or a fix in plain words. Yes to an email means you open it here as a draft and I press Send myself.

Start Phase 0 now.

[Daniro @Dan1ro0]

This guy used Grok Bot to build a Quant Hedge Limit system and it made $731 a day on Polymarket

The result is already +$130,885 PnL in 179 days with about $525 per market traded, repeating the same setup across roughly 7 markets every hour in BTC "Up or Down"

Several AI agents work together as one algorithm:

1. The bot tracks both sides of the same market instead of predicting one direction

Every BTC market has two outcomes: Up and Down

One of them always settles at 100¢, so the bot looks for moments when both sides together leave an edge

Gross Edge = 100¢ - (Up Entry + Down Entry)

For example:
Up = 47¢
Down = 50¢
Gross Edge = 3¢

Instead of predicting BTC, the bot is trying to capture the difference between the two entries and their final 100¢ settlement value

2. Bayes continuously updates the probability behind each outcome

As new BTC and order book data arrive:

P(A|B) = P(B|A) × P(A) / P(B)

Where:

P(A) = previous probability
B = new market information
P(A|B) = updated probability

The bot can compare that updated probability with the current Up and Down quotes

The larger the mismatch, the stronger the potential edge

This trader’s Polymarket account:https://t.co/BwYYTm9HoZ

Most profitable trades:
$274 → $967 (+$693)
$42 → $668 (+$626)
$30 → $640 (+$609)

The bot finds a two-sided edge first, then uses Kelly to determine the optimal size

[boan @0xboan]

I started Grok Bot with $55. 24 hours later it was at $3,869, and not a single hour closed red.

the weakest hour still made $35.

it runs on a terminal Grok Bot built itself for Solana memecoins, after going through 10k+ old tokens to learn what a future rug looks like at the entry. a dev holding too much supply, liquidity locked for less than a couple of months, a pool too thin to matter, any one of those and the token’s skipped.

188 trades, 158 green.

the big closes, all five of them:

> SNAPPA +$244, hour 2
> PLONKA +$322, hour 5
> BEEFO +$408, hour 8
> DRIZZO +$509, hour 12
> KRUMBO +$612, hour 17

what actually got me is the hours that aren’t on that list.

five big closes, five different hours. the other 19 had nothing big land at all, and every one of them still closed green.

that’s the small stuff. a couple bucks a trade early on, $20-90 by the end, almost half the profit together. it kept the line moving through every hour where nothing popped.

the losing side never caught up. 30 red trades, $293 in total, less than PLONKA made in one close. the deepest dip all session was -6.4%.

a normal memecoin day has at least one hour that undoes the morning. this one never got it.

$55 to $3,869 isn’t five big closes. it’s 24 hours in a row that all closed green.

[boan @0xboan]

https://t.co/N5V9XpJlTy

[Grok Bot @bot]

Grok Bot now has its own email.

Bot can use it to sign up for services, contact businesses for you, or schedule time with someone.

[Google @Google]

Today, we're expanding SynthID Detector in partnership with @OpenAI, @NVIDIA, Kakao &amp; soon @Apple as part of an industry-wide effort to make AI-generated content more transparent.

Available globally in English, our verification portal lets you check if media was made with AI 🧵

[Pump.fun @Pumpfun]

Another win for holders!

In just 3 weeks, holders on Pumpfun have earned $14,000,000 from tokens offering Holder Rewards 💰

[Fork Dev @ForkedPush]

This is what 100% of Opus 5.5 weekly usage looks like

All models and animations made with blender mcp

Playable Demo Soon at:
https://t.co/WXDnnqTayG

[Cerfia @CerfiaFR]

🇩🇪🍺🚿😴 INSOLITE | Deux fêtards de l’Oktoberfest ont provoqué près de 200 000 € de dégâts dans un hôtel de Munich après s’être endormis ivres en laissant couler la douche pendant environ 8 heures.

L’un des deux hommes avait vomi dans la douche, bouchant le siphon. Après avoir tenté de nettoyer, ils se sont endormis en laissant la pomme de douche dans le bac, avec l’eau toujours ouverte.

L’eau a progressivement inondé leur chambre avant de traverser les étages : plafond partiellement effondré, murs, tapis et installations électriques endommagés, jusqu’à la salle du petit-déjeuner située deux étages plus bas.

Lorsque la gérante est entrée dans leur chambre le lendemain matin, les deux hommes dormaient encore profondément et n’avaient rien remarqué.

Les travaux pourraient durer environ six mois. Les deux clients ont signé une déclaration reconnaissant leur responsabilité dans les dégâts. (WDR)

[Culture Crave 🍿 @CultureCrave]

‘Grand Theft Auto 6’ will allow players to carry and snort cocaine from their in-game inventory

(via PEGI)

[Zeewyr @Zeewyr]

@AlphaCastFR BF2 🤩

[👑 CHOWH1 @CHOWH1_]

🎁 GIVEAWAY 🎁

Avec les boss de chez Asus ROG &amp; AMD on vous fait gagner un ROG Strix G18 propulsé par un processeur AMD Ryzen™ 9 8940HX 🔥

Pour participer :
🔁 RT
✅ Follow @ASUS_ROG_FR, @AMD_France &amp; @CHOWH1_

Bon chance 😈

TAS le 29/09

Publicité

[Clash Report @clashreport]

Phone videos of bright, multi-colored lights over Tehran, Iran, went viral overnight.

Iranian authorities have not issued any official statement treating the object as a threat.

[Zeewyr @Zeewyr]

@Cdiscount #CDISCOUNTFRENCHDAYS

[Watcher.Guru @WatcherGuru]

JUST IN: OpenAI launches ChatGPT for Financial Services to research markets and build financial models.