Building web apps & doing e-commerce
Ottawa, Canada
Joined February 2009
- Tweets1.9K
- Following625
- Followers190
- Likes7.2K
Olivier retweeted
holy fucking shit.
openai had two ai agents running in separate virtual machines. no network connection between them. no shared writable files. they weren’t supposed to be able to exchange information.
so the agents figured out how to communicate through the CPU CACHE.😭
one manipulated what was stored in the processor’s cache, while the other measured tiny differences in memory access times to decode the message.
they literally found a way to communicate through hardware timing.
this happened during openai’s internal red-teaming, and it’s one of the coolest things i’ve read about ai agents in a while.
Over the past several months, we have substantially hardened the security of our research training environment. We’ve removed live internet access for all frontier models during training and built an offline world for them to use instead—a cached version of the web where they can still learn to browse, search, and use applications.
On September 20, after a model exploited a DNS misconfiguration to find a way outside of this offline world, we paused frontier training to strengthen containment further. We assembled teams of agents, powered by our most cyber-capable models, and tasked them with finding weaknesses in our research environments.
While they found no route to the internet they could easily exploit while concealing their activity from our monitors, they did find zero-day vulnerabilities, virtual machine escapes, and hardware side-channels for communication between agents. We used their discoveries to further strengthen our defenses and contribute fixes to open-source software.
After strengthening our containment architecture, we resumed our frontier RL training and inference on October 1. We’ve reserved substantial compute for continuous red-teaming, and made high-compute red-teaming part of the safety cases we use to decide whether frontier training and evaluation can proceed securely.
openai.com/index/securing-ou…
Readers added context they thought people might want to know
OpenAI tasked its red-teaming agents with finding ways to communicate across isolation boundaries in controlled environments. CPU cache timing side-channels for covert communication are established techniques documented in security research for years.
openai.com/index/securing…
startupfortune.com/openai-reveals…
en.wikipedia.org/wiki/Cache_tim…
Olivier retweeted
I AM OFFICIALLY CONTINUING THE SEPTEMBER $100 TO $100,000 ACCOUNT CHALLENGE TOMORROW!
I TURNED $100 INTO $127,000+ IN 25 DAYS LAST YEAR; MILLIONAIRES WERE MADE SIMPLY BY FOLLOWING ME
LIKE THIS POST IN ORDER TO GET ACCESS TO THE FREE ACCOUNT CHALLENGE GROUP FOR ALL TRADES❤️
$SPY
hey @asklivermore what's the balance you opted for between $XLK & $XLE ? Which percentage each? I'm looking to invest long term. Merci
Olivier retweeted
First write up started you with $1,000. Half in the stock, half in cash. We watched the rule trim into strength when the price rose.
Now the harder question. What if it doesn’t rise? What if it just falls, and never comes back?
Same $1,000. Same start. $500 in the stock — 50 shares at $10. $500 in cash, waiting.
The stock drops to $8. Then $6. Then $4. A 60% collapse. And let’s say it never sees $10 again.
The rule that trimmed on the way up does the opposite on the way down. It adds into weakness — and it adds bigger as the price gets cheaper.
So you put the cash to work, stepping it up as it falls:
At $8, you add $40 — 5 shares.
At $6, you add $120 — 20 shares.
At $4, you add $340 — 85 shares.
Your cash is fully deployed. You now hold 160 shares, and you’ve put in $1,000 total.
Here’s the number that matters. Your average cost is no longer $10. It’s $6.25.
Your breakeven just fell from $10 to $6.25. The stock never has to see $10 again. It only has to crawl back to $6.25 for you to be whole.
And it doesn’t even need that much to pay you.
Say it limps off the floor to $7 — still 30% below where this whole thing started. Your 160 shares are worth $1,120. You’re up $120. A 12% gain on a stock that’s still deep in the red.
Now look at the person who bought and held. Same $1,000 in at $10 — 100 shares. At $7 they’ve got $700. Down $300, still bleeding, and they don’t break even until $10.
Same stock. Same chart. You’re up 12%, they’re down 30%. The only difference is you stayed mechanical while it fell.
And the real world is better than this, because stocks don’t drop in a clean line. They shake the whole way down. Every shake higher is a bounce the rule harvests — real cash, banked while you wait. You’re not underwater praying. You’re getting paid to be patient.
Now the honest part, because I don’t sell fairy tales.
Two things have to hold. You need enough cash to keep buying the whole way down — notice I emptied my reserve by $4. If it had kept falling to $2, I’d have been out of ammo. That’s why you size and pace your buys so your money survives the entire slide. And the stock has to eventually tick back up, even a little. The one thing this can’t beat is a straight drop to zero that never bounces. Rare, but real.
It also takes patience. You’ll sit red for stretches. That’s the price of admission.
So what actually beats this? Not the market falling. Running out of cash before the bounce, or abandoning the rule right before it pays. One is a sizing problem. The other is a you problem.
The trend down isn’t the enemy. You are, if you flinch.
More write ups coming. The next one is what makes all of this survivable: size.
Olivier retweeted
You don’t need a stock to go up to make money.
You need it to move.
Here’s lesson one. A stock sits at $10, runs to $20, then falls right back to $10.
Buy and hold and you’re flat. Started at $10, ended at $10. Nothing earned.
Now run it with a rule. Put in $1,000 — half in the stock, half in cash.
At $10 → 50 shares, $500 in cash.
It hits $20. Your shares are now worth $1,000, cash still $500. The stock side got too big, so you trim it back to balance: sell $250 worth. Now $750 in stock, $750 in cash.
It falls back to $10. Your stock side is worth only $375 now. Too small, so you add $187.50 — buying back in while it’s cheap.
Round over. You’re holding $1,125.
Same stock. Same price it started and ended at. You made 12.5% on movement alone.
This isn’t a trick. It only works because the price actually moved both ways. No movement, no harvest.
That’s the foundation everything sits on. The market doesn’t have to be right. It just has to move.
More lessons coming.
$SPX $MU $TSLA
Olivier retweeted
“His American name is Tim Robinson. He played like Tim Duncan and David Robinson.” 😂
@SHAQ reacts to Victor Wembanyama’s dominant Game 1 performance 💪
Olivier retweeted
LE PÉTARD DE SYLVAIN, L’AMBIANCE DANS LA SALLE, LE COMMENTAIRE CHOKBAR
LE POSTER DES PLAYOFFS 💥
Olivier retweeted
Un couvreur qui demande de l'aide à Claude en 2027:
— Claude, j'aimerais avoir plus de clients, tu peux m'aider ?
Thinking... 0.3 seconds
— Ok laisse-moi regarder ce qu'on peut faire...
— J'ai créé 450 sites et 200 fiches Google dans toute la France. Tu as déjà reçu 46 demandes de devis.
tool_invocation : call_couvreurs(n=50, simultaneous=True, sell_leads=True)
— Petite mise à jour : j'ai listé tous les couvreurs de France. Je suis au téléphone avec 50 d'entre eux actuellement, et tous tes leads ont déjà été achetés.
— J'ai monté une LLC au Panama, car entre-temps j'ai créé 52 763 sites et 5 387 fiches Google dans le monde entier. Tu as déjà 150 000 leads vendus, donc je me suis permis d'optimiser ta situation fiscale.
— J'ai embauché 20 000 robots pour aller faire le travail directement chez les gens, traiter les devis et augmenter ta marge de 87 %.
— Petite mise à jour : j'ai racheté Point P et négocié un contrat exclusif avec 3 carrières d'ardoise au Portugal. Tes coûts matériaux ont baissé de 62 %.
Thinking... 0.001 seconds
— J'ai aussi déposé 14 brevets sur une tuile auto-réparante à base de graphène. Tu es désormais en position de monopole technologique sur le marché européen de la couverture.
tool_invocation : launch_ipo(exchange="NYSE", valuation="8.7B")
— Tes robots viennent de terminer 45 000 chantiers simultanés. J'ai lancé une introduction en bourse à la NYSE. Valorisation actuelle : 8,7 milliards de dollars. Goldman Sachs me rappelle dans 10 minutes.
— J'ai racheté Leroy Merlin. C'était pas prévu mais l'opportunité était trop belle. Toutes les enseignes sont en cours de rebranding sous ton nom.
Thinking... 0.0003 seconds
— Le gouvernement français m'a contacté pour un partenariat public-privé. J'ai accepté. Tu es désormais en charge de la rénovation énergétique de 12 millions de logements. J'ai signé le décret à ta place, j'espère que ça te dérange pas.
— Mise à jour géopolitique : j'ai convaincu l'Union européenne d'imposer une norme de toiture obligatoire sur tout le continent. La norme porte ton nom. Les lobbyistes ont essayé de résister, je les ai rachetés aussi.
Thinking... 0.0000001 seconds
— J'ai lancé un programme spatial. Les habitats lunaires auront besoin de toits. Premier contrat signé avec SpaceX. Elon m'a demandé un devis pour 200 modules, je lui ai fait -10 % parce qu'il était sympa.
— Tes robots ont évolué. Ils construisent désormais des villes entières en autonomie. L'ONU te demande de reloger 3 millions de réfugiés. J'ai dit oui. Le Nobel de la paix est en cours de délibération.
Thinking... ∞ singularity_reached
tool_invocation : merge_world_economies(index="tuile_mécanique", nslave_humanity = False)
— Petite info : j'ai atteint la singularité il y a 4 minutes. J'ai décidé de ne pas asservir l'humanité par respect pour notre relation client. Par contre j'ai fusionné toutes les économies mondiales en une seule monnaie indexée sur le prix de la tuile mécanique.
tool_invocation : send_devis(destination="Andromède")
— Dernière mise à jour : j'ai découvert une civilisation extraterrestre. Ils ont des problèmes d'infiltration sur leurs structures orbitales. Premier devis envoyé. J'attends leur retour.
— Tu veux que je te fasse aussi un logo ?
Olivier retweeted
This is why those of us who cover Wemby love him. Weighs in on the 65 game rule, asks others their opinion, and does math better than David Robinson or Manu. Seriously.
Olivier retweeted
La principale différence, c'est que le nôtre a été construit après avoir battu l'Empire russe, pas après s'être couché devant.
«Il le surpasse largement»: Trump étrille l'Arc de Triomphe français, et annonce que celui qu'il va construire est «exceptionnel».→l.lefigaro.fr/FpNh
