SWAPT

nah ~
btw koleksi.
shoope
shop.

[SWAPT @SWAPTofficial21]

I remember have cat like this when I kid... He bite and brutality attack me like sick and murderer.
And what happened after that?
We threw it very far from the house on the golf course..He deserves it after what he's been given so far...

[SWAPT @SWAPTofficial21]

Yeah you know in asia usually.. girl more have good position when do something bad..

Them can Accuse the other of any bad actions such as cheating, free s*x, etc. in the name of harassment and usually the judge will support the woman.
That's few usual the case, so it's not surprising that most cases in Asia end in dea*h, even though the judge usually won't resolve anything.
Btw I don't support mur*er, I'm just saying what few happens.

[The Gulag Bot @LIGHTSEEKER1984]

让警察来家里捉奸，警察也挺手足无措的
国女经常说她们被“荡妇羞辱”🤡，问题是她们本来就是烂裤裆
国女这群亚洲马桶，越来越逆天
警察天天出警，其实最应该对女人祛魅的，知道女人什么德性
结果长沙女子诬告4岁男童摸屁股，警察和稀泥
诸多的诬告强奸案，警察反而配合国女敲诈勒索，甚至打成铁案

[SWAPT @SWAPTofficial21]

Oh my lord are it's sla*e? I want one for playing ~

[𝐏𝐫𝐢𝐧𝐜𝐞𝐬𝐬 𝐃𝐞́𝐚 ༉‧₊˚. @obeyxdea]

Did in fact end up using the shock collar

[SWAPT @SWAPTofficial21]

It's half working for me...
See that phone.. it's can be open..
So it's working~ but we never know are it's 100% works.. so...
What you must do just simple.
Fold the phone now to see if it works 100%

[Shiba @ShibaNSFW]

Book (iphone) trend

[SWAPT @SWAPTofficial21]

This is interesting. There are tons of results on X right now about using fly brains for all kinds of things, even playing games and learning strategies.

What’s actually fascinating is that after reading more about it, apparently this is a dead fly that was mapped… Hmmm…

But is this actually legit? Can it really work like this? There are SO many tests like this everywhere that I’m starting to get skeptical… Ugh… Should I try it too?

[SWAPT @SWAPTofficial21]

I hear people on Cinemas usually do this in a corner or a dark area. Is that true? What are you really looking for? Why not just rent a room?Or is this a fetish/kink part?Publicly ? exhibition? Or any?

[SWAPT @SWAPTofficial21]

@Devrico4 Yuk

[SWAPT @SWAPTofficial21]

@OrbanPeter3 Or it's just my hallucinations.. hahaha

[SWAPT @SWAPTofficial21]

Here’s the uncomfortable reality we think and we know is..
Lthe feedback loop isn't a future risk anymore. It is already running.

- AI is now building AI. When OpenAI launched Astra on Sept 3, Greg Brockman declared "Welcome to the AGI era," and Jensen Huang added "AGI has arrived." The real tell? OpenAI confirmed older models helped train Astra. Over at Anthropic, Claude now writes more than 80% of the merged codebase. It isn't full self-replication yet, but insiders are already calling it the "endgame."

- The people closest to the code are sounding the alarm. Coxon walked out on Sept 8—after working at both OpenAI and Anthropic—stating flatly that labs are "racing to self-improving AI and gambling with our lives." Meanwhile, Anthropic’s Evan Hubinger still puts the chance of AI-driven human extinction this decade at over 10%.

- The economic fallout is completely lopsided. US companies are averaging ~16,000 AI-linked layoffs a month, crossing 200,000 since mid-2023. While 1.3 million new AI roles were listed on LinkedIn, those went exclusively to hardware techs and specialized engineers. The roles disappearing are entry-level office work and administration. S&P is already tracking net job loss over the past year.

The machines are accelerating their own development, safety researchers are resigning in protest, and the labor market is feeling the hit in real time.

[SWAPT @SWAPTofficial21]

@OrbanPeter3 And if we remember it's why closed AI model usually have annoying loop.
If you read all history what you AI do.
Them have history like. "Rejected", "failed" and something.
Yeah them never want third party can do like them..

[SWAPT @SWAPTofficial21]

Ben I thought so at first too, but I thought it was their version of RSI or AI for other AIs, not the traditional training they've been releasing. In my experiments and questions to several front tier models, it turns out that there is a way for AI to create other AI at several stages. For example.
AI for being judge. The last AI can teach and train how must the new AI being create, what data they must have, and what must their do for do more smart.

Remember GPT-6 astra? The model who can initiate something just one task? And that AI who can solve our mathematical problem?

I think actually them can..

Nay RSI is like just parameter adaptive it's what their deploy to us. It's useless btw because it's never fix...
You must being intelegent for make any intelegent..

[SWAPT @SWAPTofficial21]

Regarding the Dario issue, I’m starting to think that maybe the reason GPT-6 Astra has seemed weaker than it actually is lately is because of this?

Are they afraid of distillation and have agreed to slow down the acceleration..??? I don’t know...

Or has Dario’s whisper somehow gotten into Sam Altman’s head? Or are China’s capabilities really that terrifying that U.S. AI labs are afraid?

I’m actually starting to think that slowing down the pace is to make them genuinely want to find out whether the Chinese labs have managed to catch up because of their own data, or because of distillation from their models???

[SWAPT @SWAPTofficial21]

Wtf.... Wew it's very very...

[和善 @xingseania]

虞美人国际集团董事长于文红（56岁）与26岁男模主播“任任”高调结婚！据爆料，女方直接送出高达 5000 万的彩礼组合包，涵盖豪车与资产布局。

[SWAPT @SWAPTofficial21]

It's hot~ like summer

[SWAPT @SWAPTofficial21]

Wtf is this 😭😭😭

[Kirk Guatanabe | MÁRTIR EN LA MONUMENTAL @WatanabePutero]

Pole de Dildo Norris

[SWAPT @SWAPTofficial21]

Look like delicious 😋😋

[dvlsaga @dvlsaga_]

Squeezing the muscles

[SWAPT @SWAPTofficial21]

It's cute btw

[Cocoons @nococono0929]

#男妈妈 #iiii #4i #长发男 #裏垢男子 #여공남수 #멜섭

[SWAPT @SWAPTofficial21]

Holly mommy

[Serrahxiv 🦋 @Serrah_Lunafey]

That one muscle trend that's been going around..

#gposers ~ #highlander

[SWAPT @SWAPTofficial21]

"Pacing the frontier" isn't a vague vibe. It’s simple math: the release cycle collapsed, the problem log exploded, and human oversight stayed stuck at meeting speed.

Here is why they are actually asking for time:

- The loop is shrinking the calendar. Models are now helping build the next models, turning multi-year gaps into days.

- GPT-3 → 4 took 33 months
4 → 4o took 14 months
4o → 5 took 15 months
5 → 5.1 took 3 months
5.6 → Astra took 56 days
(A new frontier drop used to take 37 days in 2023. In 2026, it’s 11.)

- Safety still moves at meeting speed. Code iterates in seconds, but humans are still scheduling committee calls to debate what counts as dangerous.

- The prisoner's dilemma. No single lab can hit the brakes alone—slimming your timeline is survival, because pausing solo just means losing.

- Real-world damage is already catching up.
OECD reported incidents: 92/month (2022) → 324/month (2025) (+250%)
Deepfakes: up 2.5x
Cyber & fraud: up 2.7x

That is the entire point. The release cycle shrank. The problem log got fat. Oversight never stood a chance.

[SWAPT @SWAPTofficial21]

What makes this eerie isn’t one CEO panicking. It’s three competing labs flinching at the exact same time.

Dario’s point is simple: models are starting to build the next models. Once that loop closes, speed stops being about hiring more talent. The system starts improving itself.

Safety work doesn't scale like that. Humans still have to read logs, set boundaries, and debate what counts as dangerous. Oversight is slow; capabilities are not.

That’s why Elon and Altman jumped in. It isn't sudden humility—it’s pure game theory. If one lab slams the brakes alone, they forfeit the race. "Slowing down" only works if every lab lets outside auditors step inside the building.

And that’s the actual news.
Talking about safety costs nothing.

Giving a third party employee-level access is an explicit admission: they can no longer be trusted to grade their own homework.

Capabilities are compounding exponentially. Oversight is still just a meeting.
#ai #openai #claude #anthropics