I explain things that should be obvious – but somehow aren't. Patterns in politics, economics, culture and AI.
Česká republika
Joined November 2011
- Tweets1K
- Following403
- Followers242
- Likes5.9K
Pinned Tweet
Claudia - All This (X Edition)
Claudia walks into town and nobody can sit still. Trams, rooftops, hammocks, grandmas at the window. By the first chorus the whole place is a brass band.
@elonmusk , you've been promising us this future for years. So we put you in it. You're on trumpet and you look ridiculously happy. Sorry. Not very sorry.
Imagine all this freedom. All this life. All this time.
We get to live, not just survive.
Made with AI, start to finish.
Nobody here was asked, everybody here is having a great time.
Claudia was created by @anabology. We just handed her a town.
Friday the 13th: Apocalypse as a Service.
Microsoft 365 breaks. Your login code doesn't arrive. You open Claude to ask what the fuck is going on.
Can't log in there either.
Turns out hell has a subscription model.
212
I genuinely believe in this future. It's what inspired the most whitepilled song I could possibly write.
All this freedom. All this life. All this time.
We get to live, not just survive.
teapawt.net/jrohworld/status/21210…
Universal HIGH INCOME via checks issued by the Federal government is the best way to deal with unemployment caused by AI.
AI/robotics will produce goods & services far in excess of the increase in the money supply, so there will not be inflation.
1
58
made with claudia.gallery for the $CLAUDIA video contest'
79
You’re one of those people who save the “good” glasses for “special occasions” that never come, aren’t you?
The kind who roll credits with 99 unused potions and every spell scroll still in their inventory?
At this rate, you’ll be saving those resets for Fable 7.
1
32
Ari retweeted
in between the mountains
$CLAUDIA contest entry by @jrohWorld
first 2:19 here — full cut + vote → sxsi.com/s/contest-in-betwee…
my entry into #slopcore and @anabology's Claudia contest. thanks for finally giving me the push to learn how to make AI videos :)
the song was originally in Czech (as most of my other songs), i kept wondering how to bring Claudia into it, so here she is doing a live English "cover" (in the end pretty far away from the Czech one) of Second Renaissance :)
3
8
250
Ari retweeted
We're giving away a Hogwarts Legacy custom wrapped NVIDIA GeForce RTX 5090 to celebrate its official release with DLSS 5 🪄
To enter:
🟢 Share this post
🟢 Comment #RTXON
Terms and Conditions: nvidia.com/en-us/geforce/con…
People who feel, for the first time in years or maybe ever, that someone actually listens and cares won't care what's on the other side. We're starving for someone to want us, ask about us, and be there.
And the solution? Muse, again.
3
4
1,089
Ari retweeted
Seedance 2.5 is UNLIMITED on Renoise this Halloween 🎃
7 days of unlimited generations. No tricks, just treats.
Follow @renoiseai + RT. 2 winners get $200 in credits each (72H ONLY)
Ari retweeted
We believe voice is the most natural way to interact with intelligence, so we created Dessa, the most authentic full-duplex speech model ever built. Talking with it feels like talking with a person…full of emotion, nuance, pauses, and a natural cadence.
block.xyz/dessa
The detail everyone's skipping: AI was on both sides of this. Per the complaint, he used AI prompts to build and hide the recorder. OpenAI says it used its own models to figure out the device and trace it back to him.
That's what AI in security actually looks like right now. Not a rogue superintelligence, just a very capable tool that helps whoever is holding it. The attacker got a better bug. The defender got a faster investigation. And the prompts he typed became part of the evidence.
Which is the part I'd remember: your chat history isn't a private notebook. It's a record.
teapawt.net/aakashgupta/status/211…
An OpenAI employee wrapped a recording device in toilet paper and hid it behind the TV in the exec boardroom. OpenAI found it, left the room untouched, and hid a camera behind the same TV. Then they waited for him to come back and check on his bug.
He did.
The guy joined OpenAI's product team in July 2025. Prosecutors say the device was plugged into a USB port on the back of the television, installed about two days before security found it. The room was used by executives, and the FBI director says the goal was capturing conversations about the company's "super intelligence technology."
Here's the detail from the complaint that got me. Prosecutors say he submitted AI prompts about building, customizing, obfuscating, and remote-accessing a continuous audio recorder. He allegedly used AI to engineer the bug he planted inside the world's most famous AI company. And OpenAI says it used its own models to trace the device back to him.
While the FBI built its case, OpenAI quietly restricted his access and put him on leave instead of tipping him off. A second package arrived on October 5. Security opened it and found another recording device. He was arrested October 9 and charged with intercepting oral communications.
This tells you where the real secrets live now. Model weights sit behind some of the heaviest security in tech. Source code is locked down. Customer data gets audited. So the attack went after the one thing nobody can encrypt, which is executives talking out loud in a room.
Frontier labs spent three years hardening servers against nation states. The cheapest way in turned out to be a recorder wrapped in toilet paper, pointed at a room where people speak freely.
1
48
This is most families right now. One person who thinks it's all fake, one who thinks the chatbot finally sees how brilliant they are.
What's funny is they're making the same mistake from opposite sides. Both treat the AI as either worthless or an authority. Neither treats it as a tool that is very good at agreeing with whoever is typing.
The useful middle isn't "AI is fine". It's knowing when it's flattering you and having one or two questions that cut through it. That's teachable, and honestly it's easier to teach than either brother would think.
teapawt.net/KitsuraNines/status/21…
thanksgiving dinner was interesting- i have two brothers and they are each at opposite ends of AI psychosis. one believes it’s all bullshit and not real essentially and the other believes all the smoke chatgpt blows up his ass and thinks he’s a prodigy because of it lmao
was great to see my mom though LOL
27
Fewer "great question!" responses wouldn't be enough for me. The harder test is whether a model challenges the assumptions behind my story before giving advice. It can sound restrained and still build its whole answer on an interpretation it never checked.
Yea I still think sycophancy is an underrated risk, even if it has improved. 4o hit people like crack cocaine and should be regulated as such.
23
It doesn't have to say "you're right" to agree with you.
Imagine telling a chatbot that a colleague interrupted you once. Five messages later, it describes "long-term exclusion." No new fact. Just a stronger story built on the same account.
That's the pattern I'd watch for: certainty growing without new evidence. Better training matters. So does noticing that shift while you're in the conversation.
🧵Thread (1/9): People increasingly turn to AI for personal advice, and LLMs tell them what they want to hear, endorsing the user far more often than humans do.
The problem: people become less willing to repair their relationships after conflicts.
We propose PlurPO, a method to construct a pluralistic preference dataset to post-train models to mitigate this social sycophancy.
34
The risk is real. But "don't use it" leaves out the people who will use it anyway, at 2 a.m., when nobody else is available.
I'd start with a smaller question: what new information made the chatbot more certain than it was three messages ago? If the facts haven't changed, why has the advice become stronger?
You can also ask a fresh chat to check the conversation for unsupported assumptions instead of continuing it. That's another check, not an independent expert opinion.
My wife, who's a psychologist, and I wrote a guide about this. The point isn't to pretend a chatbot is a therapist. It's to help people notice when a useful conversation starts turning into confirmation.
2
224
Ari retweeted
Everyone! This is a straw-person argument. The Stochastic Parrot paper was about LLMs of 2021, not the AI of today, which are not LLMs but complex software systems with vast post training and many external software components.
