designing interfaces, testing agents, occasionally posting nonsense. don’t take me too seriously
Joined August 2026
- Tweets4.9K
- Following321
- Followers433
- Likes1.9K
Pinned Tweet
Day 6 of diving deeper into cybersecurity
Monitoring & AI watchers (telemetry, records of what your system did, watched by rules and by AI, so you can catch an attack while it is happening)
You cannot stop what you cannot see. And seeing it late is almost as bad as not seeing it at all. An attacker who has been in your system for weeks has already had time to learn, move, and take what they want.
A watcher is not magic. It is telemetry, plus triage, plus a real response. Here is what goes unnoticed, and how to fight each one:
1. Blind: no logs, no view.
If you do not record logins, code changes, new admins, large exports, and key use from new places, you will not know an attack happened until someone else tells you.
Fight it with telemetry. Telemetry just means records of what happened. Log what matters, keep those logs somewhere the attacker cannot delete them, and make sure you can actually read them later.
2. Noisy: real signal buried.
If everything alerts, nothing alerts. People start ignoring alerts, and the real one gets missed in the noise.
Fight it with triage. Triage means sorting what matters from what does not. This is where AI helps best, as a sorter, not as a decider. AI can summarize a flood of logs and point to the odd ones, and a human or a rule decides what to do about them.
3. Late: found weeks later.
A slow discovery gives the attacker time to finish the chain from Day 1.
Fight it with fast alerts. Alert in minutes, not weeks, on the critical things: code or config changes to auth, payments, or signing, a new admin, a failed login spike, a large export, a key used from a new place. Those are the things worth waking up for.
4. No context: alert, but who, what?
An alert that just says "something odd happened" does not help you act.
Fight it with rich context. Every alert should say who, what, where, and when, in plain language. Who logged in, what changed, from where, and what it touched. Without that, you waste the fast part investigating.
5. No action: seen, but not stopped.
This was the Hugging Face lesson in July 2026. Their watcher was LLM-based anomaly detection, LLM triage over security telemetry to separate real signals from daily noise, and it did flag the intrusion. But a watcher is detection, not prevention. The agents had already abused remote code and template paths, harvested credentials, and moved laterally before it was contained.
Fight it with policy action. Behind the watcher, have a rule or a human who can pause, revoke, roll back, and rebuild. An alert should trigger a response, not just a notification.
A watcher that only watches is not enough.
It has to see, explain, and trigger a real response.
ChatGPT says this is the kind of supreme dictator I’d be 👇🏻
AND THAT IT WOULD VOTE FOR ME
But would you??
Instead of individuals giving up their little bits of RAM, why don’t we require it of big companies with terrible AI like Apple?
Plot twist:
This is actually @poteto
Now being a double agent for both OpenAI and xAI
Some news: I’m joining OpenAI to lead product management for Core Products!
I have a lot of respect for OpenAI for showing the world that scaling works, for bringing AI to over a billion people, and for believing great products are core to the mission.
I’m excited to work with @thsottiaux, an incredible PM team, and everyone across OpenAI. Together, we’ll build agents that have your back and your team’s, and keep you in control, so you get more time for the people and work you love.
Lots to learn, and lots to build. Please keep your feedback coming!
And then we proceed to centralize and regulate our only moat
After DOGE, Elon came to a conclusion most people missed.
He tried to reform the state from the inside. He failed. And he said it plainly: the government can't fix its own deficit. Nobody in power actually wants to cut. The machine protects itself.
So he drew the only logical conclusion. If the state can't be reformed, there are only two ways this ends.
Either global bankruptcy. Debt keeps compounding, interest payments eat the budget, and the system collapses under its own weight.
Or robots. AI and robotics grow the economy so fast that the debt becomes manageable, and abundance makes the old fights over scarcity irrelevant.
Here's the part almost nobody sees.
For two centuries, capitalism has had a marketing problem. Its gains are real but slow and invisible. 2% growth a year doesn't feel like anything. Your life gets better over 30 years and you never connect it to the system that did it. Meanwhile, the left's promises feel immediate and concrete: a check, a subsidy, a benefit. That's why the left wins on promises even when it can't deliver them. Capitalism produces the wealth, socialism takes the credit.
Robots break that asymmetry.
When productivity doubles in a few years instead of a few decades, people will see it. Things get cheaper, services get better, opportunities appear, in real time. For the first time in history, people will connect capitalism and their rising standard of living directly, almost instantly. The causal chain becomes visible.
And that changes everything politically.
So there are two scenarios.
Scenario one: we stay in the current inertia. No breakthrough, or a breakthrough we regulate to death. The debt wins. Living standards stagnate, then fall. A frustrated electorate turns to whoever promises the most, and that pushes toward a hard collectivist left. But collectivism fails, as it always does. And when it fails, the most organized, most disciplined collectivist force left standing isn't the old left. It's radical Islamism. Bankruptcy, then communism, then something worse.
Scenario two: robots win. Abundance arrives, and it arrives visibly. The link between free enterprise and prosperity becomes obvious to everyone. And this is the key point: it becomes obvious to everyone, including the minorities the left was counting on. Because their long-term incentive is the same as everyone else's. A kid in the suburbs gains infinitely more from a booming, abundant West than from a bankrupt state handing out promises it can't keep. Abundance is the most powerful integration machine ever built.
That's the Western renaissance.
The West doesn't need to win the culture war by arguing. It needs to win it by delivering. And for the first time, it has the technology to deliver fast enough for people to notice.
The race isn't left vs right anymore. It's robots vs bankruptcy.
Omg almost to 430 followers!!!
I was stuck below 425 for soooo long
Like why doesn’t X appreciate me more??
Day 5 of diving deeper into cybersecurity
Sandboxing (running code you do not trust in an isolated, disposable place)
AI agents, downloaded code, uploaded files, and datasets all have one thing in common. You do not fully control what they will try to do. If you run them like normal code, they run with your power, your files, and your secrets.
A sandbox is a box you put them in first. It is separate, it is empty, and you can throw it away.
Here is what goes wrong when you do not sandbox, and how to fight each one:
1. Untrusted code: runs with your power.
By default, code you run can do what you can do.
Fight it with isolate. Run it in a separate box, not on your main machine and not next to production. If it misbehaves, it is contained.
2. Full access: reads files, secrets.
If the box has your credentials and your private files, the code can take them.
Fight it with no prod secrets. Give the sandbox an empty environment. No real keys, no real customer data, no access to production systems. Use fake data for testing.
3. Internet: phones home, downloads more.
With general internet, bad code can send data out and pull in more bad code.
Fight it with no internet, allowlist only. Turn internet off by default. If it truly needs a connection, allow only the exact addresses it needs, and nothing else.
4. Persistent: changes stick around.
If the same environment keeps running, a compromise can hide, spread, and come back.
Fight it with ephemeral. Ephemeral means short-lived and disposable. Start fresh for each run, and throw the box away after. Nothing should survive from one run to the next unless you meant it to.
5. No limits: runs forever, eats all.
Untrusted code can loop, fill the disk, or use all the CPU and memory.
Fight it with limits + timeout. Set hard limits on time, memory, disk, and what files it can touch, and kill it when it hits them.
Do not trust the code. Contain it.
If it escapes, it should find nothing worth taking.
This SHOWS how MANY people would be INTERESTED in a SEXBOT
Today @maryamyy0 and I are coming out of stealth with Sexbot.
A companion that's always there for you.
Built to fix the growing loneliness in the world.
We raised $5M to build it. Pre-orders are open now.
Link in the comments.
Readers added context they thought people might want to know
This is not a real product/service. It's made up to advertise this person's real company which creates launch videos.
Source: sexbot.services/kidding
