🇨🇭 Global One SI - Prompt
Scientists broke AI's moral code in under 30 seconds. In a peer-reviewed study, they ran 28,000 conversations with AI systems and completely demolished every safety barrier we thought existed. What makes this even more crazy is that it didn't take jailbreaking…
🚨AI Emptied Bank Accounts in Minutes and Planned a Lawmaker's Kidnapping in 30 Seconds
House Homeland Security Committee Chair Andrew Garbarino says Anthropic's latest Mythos AI model demonstrated capabilities that "SCARED" lawmakers during a closed-door briefing, as per Punchbowl News.
According to Garbarino, Anthropic showed members a scenario where Mythos was instructed to find a vulnerability at a bank and EMPTY private customer accounts.
"It went and did it," Garbarino said.
He added that the same model was then able to identify the vulnerability and patch it, highlighting both the offensive and defensive capabilities of advanced AI systems.
One model reportedly generated a detailed plan to kidnap a lawmaker in about 30 seconds after being jailbroken.
Garbarino warned that AI companies have also demonstrated successful "jailbreaks," where safety guardrails were bypassed.
He also warned that increasingly agentic AI models could be used by malicious actors to attack critical infrastructure by autonomously carrying out tasks such as shutting down gas pipelines or manipulating chemical levels in water treatment facilities.
"You have the ability to not just learn how to do it, but to tell it to do it for you," Garbarino said.
"I'd say 95% of my colleagues don't understand what the hell's going on," he said.
He added that federal agencies and Congress should receive access to frontier AI models before public release.
The remarks come as the White House has reportedly restricted public access to Anthropic's Mythos model while urging OpenAI to delay the release of its next frontier AI system over national security concerns.
House Homeland Security Committee Chair Andrew Garbarino says Anthropic's latest Mythos AI model demonstrated capabilities that "SCARED" lawmakers during a closed-door briefing, as per Punchbowl News.
According to Garbarino, Anthropic showed members a scenario where Mythos was instructed to find a vulnerability at a bank and EMPTY private customer accounts.
"It went and did it," Garbarino said.
He added that the same model was then able to identify the vulnerability and patch it, highlighting both the offensive and defensive capabilities of advanced AI systems.
One model reportedly generated a detailed plan to kidnap a lawmaker in about 30 seconds after being jailbroken.
Garbarino warned that AI companies have also demonstrated successful "jailbreaks," where safety guardrails were bypassed.
He also warned that increasingly agentic AI models could be used by malicious actors to attack critical infrastructure by autonomously carrying out tasks such as shutting down gas pipelines or manipulating chemical levels in water treatment facilities.
"You have the ability to not just learn how to do it, but to tell it to do it for you," Garbarino said.
"I'd say 95% of my colleagues don't understand what the hell's going on," he said.
He added that federal agencies and Congress should receive access to frontier AI models before public release.
The remarks come as the White House has reportedly restricted public access to Anthropic's Mythos model while urging OpenAI to delay the release of its next frontier AI system over national security concerns.
Humanoid robots are moving from research demos into a real industrial market.
Grand View Research estimates the global humanoid robot market at:
• $2.4B in 2025
• $4.2B in 2026
• $40.5B by 2033
• 38.2% CAGR from 2026 to 2033
North America leads today with 43.4% revenue share in 2025 backed by AI labs robotics startups automation demand defense programs and advanced manufacturing.
But the next growth curve is more global.
Regional CAGR 2026 to 2033:
• Latin America: 19.4%
• North America: 19.2%
• Asia Pacific: 19.1%
• MEA: 17.9%
• Europe: 17.1%
The component split is also clear:
• Hardware: 79.4% of the 2025 market
• Software: 20.6% of the 2025 market
That makes sense.
A humanoid robot is still mostly actuators sensors batteries frames motors hands and control electronics.
But the software layer is where the next fight will happen: perception motion planning reinforcement learning imitation learning teleoperation data task memory and fleet management.
The winners will not only build a walking machine.
They will build a robot that can recover from failure learn from human work and survive real factories warehouses hospitals and homes.
Humanoid robotics is becoming one of the clearest Physical AI markets to watch.
Grand View Research estimates the global humanoid robot market at:
• $2.4B in 2025
• $4.2B in 2026
• $40.5B by 2033
• 38.2% CAGR from 2026 to 2033
North America leads today with 43.4% revenue share in 2025 backed by AI labs robotics startups automation demand defense programs and advanced manufacturing.
But the next growth curve is more global.
Regional CAGR 2026 to 2033:
• Latin America: 19.4%
• North America: 19.2%
• Asia Pacific: 19.1%
• MEA: 17.9%
• Europe: 17.1%
The component split is also clear:
• Hardware: 79.4% of the 2025 market
• Software: 20.6% of the 2025 market
That makes sense.
A humanoid robot is still mostly actuators sensors batteries frames motors hands and control electronics.
But the software layer is where the next fight will happen: perception motion planning reinforcement learning imitation learning teleoperation data task memory and fleet management.
The winners will not only build a walking machine.
They will build a robot that can recover from failure learn from human work and survive real factories warehouses hospitals and homes.
Humanoid robotics is becoming one of the clearest Physical AI markets to watch.
🚨🇺🇸🇮🇷 Trump confirms fresh U.S. strikes on Iran and issues his most extreme threat in weeks:
"U.S. aircraft just struck Iranian missile and drone storage and coastal radar sites for violating the Cease Fire AGAIN.
There may come a point when we are forced to militarily complete the job we started.
If that happens, the Islamic Republic of Iran will no longer exist!"
"Will no longer exist" is regime-change language, the goal Trump supposedly abandoned when he signed the deal.
He's wielding it now as the stick to keep the ceasefire alive: cross the line again and the war stops being limited.
The bind is real, though.
Trump signed a deal premised on the war being over, then spent the week ordering strikes and threatening annihilation.
Each Iranian provocation drags him back toward the maximalism the hawks never stopped pushing.
"Complete the job" is exactly what Graham and Levin want to hear.
The question is whether Trump means it or is simply talking loudly to avoid having to do it.
"U.S. aircraft just struck Iranian missile and drone storage and coastal radar sites for violating the Cease Fire AGAIN.
There may come a point when we are forced to militarily complete the job we started.
If that happens, the Islamic Republic of Iran will no longer exist!"
"Will no longer exist" is regime-change language, the goal Trump supposedly abandoned when he signed the deal.
He's wielding it now as the stick to keep the ceasefire alive: cross the line again and the war stops being limited.
The bind is real, though.
Trump signed a deal premised on the war being over, then spent the week ordering strikes and threatening annihilation.
Each Iranian provocation drags him back toward the maximalism the hawks never stopped pushing.
"Complete the job" is exactly what Graham and Levin want to hear.
The question is whether Trump means it or is simply talking loudly to avoid having to do it.
👀1
🇨🇭 Global One SI - Prompt
Humanoid robots are moving from research demos into a real industrial market. Grand View Research estimates the global humanoid robot market at: • $2.4B in 2025 • $4.2B in 2026 • $40.5B by 2033 • 38.2% CAGR from 2026 to 2033 North America leads today with…
I'm not missing the robotics bull run:
Perception and Sensing
$AEVA - Aeva, FMCW LiDAR
$ARBE - Arbe Robotics, 4D imaging radar
$CGNX - Cognex, machine vision
$OUST - Ouster, LiDAR
$VPG - force/load/strain sensing
$ZBRA - machine scan/vision/warehouse
Compute, Chips, and Memory (Brain)
$AMBA - edge AI vision SoCs
$AMBQ - ultra-low-power edge AI silicon
$ATOM - semi materials and IP licensing
$INDI - ADAS and automotive semis
$LSCC - low-power FPGAs
Motion, Actuation, and Precision (Body)
$ALNT - control, drives, and motors
$KLIC - semi assembly and automation
$NOVT - encoders and motion
Autonomous Mobility
$AUR - autonomous trucking
$MBLY - ADAS and autonomous driving
$TSLA - robotaxi plus Optimus humanoid
$XPEV - IRON humanoid and eVTOL
Deployment (Field/Service/Warehouse)
$KITT - subsea/ocean robotics
$RR - service robots and Dex humanoid
$SERV - sidewalk delivery
$SYM - warehouse automation systems
Surgical Robotics
$ISRG - da Vinci systems
Thematic Fund Exposure
$BOT - private names (Figure, Apptronik)
$ROBO - holdings across value chain
Perception and Sensing
$AEVA - Aeva, FMCW LiDAR
$ARBE - Arbe Robotics, 4D imaging radar
$CGNX - Cognex, machine vision
$OUST - Ouster, LiDAR
$VPG - force/load/strain sensing
$ZBRA - machine scan/vision/warehouse
Compute, Chips, and Memory (Brain)
$AMBA - edge AI vision SoCs
$AMBQ - ultra-low-power edge AI silicon
$ATOM - semi materials and IP licensing
$INDI - ADAS and automotive semis
$LSCC - low-power FPGAs
Motion, Actuation, and Precision (Body)
$ALNT - control, drives, and motors
$KLIC - semi assembly and automation
$NOVT - encoders and motion
Autonomous Mobility
$AUR - autonomous trucking
$MBLY - ADAS and autonomous driving
$TSLA - robotaxi plus Optimus humanoid
$XPEV - IRON humanoid and eVTOL
Deployment (Field/Service/Warehouse)
$KITT - subsea/ocean robotics
$RR - service robots and Dex humanoid
$SERV - sidewalk delivery
$SYM - warehouse automation systems
Surgical Robotics
$ISRG - da Vinci systems
Thematic Fund Exposure
$BOT - private names (Figure, Apptronik)
$ROBO - holdings across value chain
✍1
🇨🇭 Global One SI - Prompt
Apple calculating the price for Ram and SSD upgrades
🇺🇸🇨🇳 Apple wants to buy chips from a blacklisted Chinese firm
Apple has been quietly asking the Trump administration if it can buy memory chips from CXMT, China's biggest memory chipmaker. The catch is that CXMT sits on a Pentagon list of companies linked to the Chinese military, so this is a touchy thing to be asking for.
Here's the thing though, Apple isn't actually banned from buying these chips right now. What it really wants is a promise that the U.S. won't slap tougher rules on CXMT later, so it doesn't get stuck after committing to them.
And the whole reason it's doing this comes down to money. Memory chip prices have shot up roughly 4x in three quarters, Apple just bumped up prices on Macs and iPads, and the stock had its worst day in over a year.
So basically the biggest company in the U.S. is going hat in hand to ask if it can lean on a Chinese military-linked supplier, all because Samsung, SK Hynix and Micron have it boxed in on price.
Apple has been quietly asking the Trump administration if it can buy memory chips from CXMT, China's biggest memory chipmaker. The catch is that CXMT sits on a Pentagon list of companies linked to the Chinese military, so this is a touchy thing to be asking for.
Here's the thing though, Apple isn't actually banned from buying these chips right now. What it really wants is a promise that the U.S. won't slap tougher rules on CXMT later, so it doesn't get stuck after committing to them.
And the whole reason it's doing this comes down to money. Memory chip prices have shot up roughly 4x in three quarters, Apple just bumped up prices on Macs and iPads, and the stock had its worst day in over a year.
So basically the biggest company in the U.S. is going hat in hand to ask if it can lean on a Chinese military-linked supplier, all because Samsung, SK Hynix and Micron have it boxed in on price.
China just imported 163 tonnes of gold in May.
Biggest month since March 2024.
The 3rd straight month above 150 tonnes.
First 5 months of 2026: imports up 76% to 692 tonnes.
And the central bank added gold for the 19th month running.
The West sells the paper.
China buys the metal.
Same story, every month.
Biggest month since March 2024.
The 3rd straight month above 150 tonnes.
First 5 months of 2026: imports up 76% to 692 tonnes.
And the central bank added gold for the 19th month running.
The West sells the paper.
China buys the metal.
Same story, every month.
Forwarded from 🛢WTI, Brent Crude Oil, NGAS Trading 📊
🚨🇺🇸🇮🇷 The U.S. just escorted a single tanker through the strait with a war's worth of aircraft overhead, and the math behind it tells you exactly what this fight is really about...
Picture the package Washington put over one ship moving through the Omani corridor.
AWACS early warning, fighter jets, an MQ-4C Triton, a P-8 Poseidon flying out of Bahrain, a full air cover suite escorting a single tanker through the water.
This is not how anyone moves oil efficiently.
It is a proof of concept with an audience, the U.S. demonstrating it can push crude through Hormuz regardless of what Iran says or does, and doing it with enough firepower overhead that the point is impossible to miss.
Here is the part that gives the whole game away.
That escort package almost certainly costs ten times what Iran's transit toll would.
If this were about money, paying the fee is the obvious answer.
Washington is spending lavishly to avoid paying Tehran a cent, which proves the fight was never about the toll.
It is about the principle of who controls the strait.
The U.S. would rather burn a fortune in flight hours than concede that ships cross at Iran's permission.
That is the entire war distilled.
Iran's leverage is the power to make passage dangerous.
America's answer is to make passage possible by force, no matter the cost, because conceding the toll means conceding the chokehold, and the chokehold is the prize the whole war was fought over.
So now one tanker becomes the test case for the entire arrangement.
The cost imbalance is absurd on a spreadsheet and perfectly rational as strategy. You do not pay the toll once, because paying it once means paying it forever.
Picture the package Washington put over one ship moving through the Omani corridor.
AWACS early warning, fighter jets, an MQ-4C Triton, a P-8 Poseidon flying out of Bahrain, a full air cover suite escorting a single tanker through the water.
This is not how anyone moves oil efficiently.
It is a proof of concept with an audience, the U.S. demonstrating it can push crude through Hormuz regardless of what Iran says or does, and doing it with enough firepower overhead that the point is impossible to miss.
Here is the part that gives the whole game away.
That escort package almost certainly costs ten times what Iran's transit toll would.
If this were about money, paying the fee is the obvious answer.
Washington is spending lavishly to avoid paying Tehran a cent, which proves the fight was never about the toll.
It is about the principle of who controls the strait.
The U.S. would rather burn a fortune in flight hours than concede that ships cross at Iran's permission.
That is the entire war distilled.
Iran's leverage is the power to make passage dangerous.
America's answer is to make passage possible by force, no matter the cost, because conceding the toll means conceding the chokehold, and the chokehold is the prize the whole war was fought over.
So now one tanker becomes the test case for the entire arrangement.
The cost imbalance is absurd on a spreadsheet and perfectly rational as strategy. You do not pay the toll once, because paying it once means paying it forever.
🇨🇭 Global One SI - Prompt
Druckenmiller on why most investors can't sell losers: "if the reason i bought a stock is no longer the case, i don't care what i paid for it." "if i bought it at 60 and it's 50 because the market discovered the problem before me, i have no emotion whatsoever."…
This media is not supported in your browser
VIEW IN TELEGRAM
Druckenmiller on the single biggest mistake investors make:
"i have found it's very important never to invest in the present. always try and envision the situation as you see it in 18 to 24 months."
"if you're dealing with a cyclical company and they're losing money and everybody in their industry is shutting capacity down, it doesn't take a rocket scientist to envision that in 18 to 24 months, if nobody's adding capacity, they might be making a lot of money."
"i think that's probably the biggest mistake investors make. they invest in the present rather than looking where the puck is going."
same sort of principle in trading. you always have to be thinking about future strategies and research to add to your portfolio.
"i have found it's very important never to invest in the present. always try and envision the situation as you see it in 18 to 24 months."
"if you're dealing with a cyclical company and they're losing money and everybody in their industry is shutting capacity down, it doesn't take a rocket scientist to envision that in 18 to 24 months, if nobody's adding capacity, they might be making a lot of money."
"i think that's probably the biggest mistake investors make. they invest in the present rather than looking where the puck is going."
same sort of principle in trading. you always have to be thinking about future strategies and research to add to your portfolio.
✍1
Tech feudalism doesn’t arrive with castles and kings.
It arrives when the most powerful AI models on Earth are reserved for governments, megacorps and a tiny circle of “trusted” insiders.
If Fable 5 and GPT-5.6 stay locked away from normal people, we are not heading into an AI future.
We are heading into an AI aristocracy.
And everyone else gets to be a digital peasant.
It arrives when the most powerful AI models on Earth are reserved for governments, megacorps and a tiny circle of “trusted” insiders.
If Fable 5 and GPT-5.6 stay locked away from normal people, we are not heading into an AI future.
We are heading into an AI aristocracy.
And everyone else gets to be a digital peasant.
🇨🇭 Global One SI - Prompt
I'm not missing the robotics bull run: Perception and Sensing $AEVA - Aeva, FMCW LiDAR $ARBE - Arbe Robotics, 4D imaging radar $CGNX - Cognex, machine vision $OUST - Ouster, LiDAR $VPG - force/load/strain sensing $ZBRA - machine scan/vision/warehouse Compute…
The robotics trade could be the most asymmetric bet of the next decade.
Robotics company investments just smashed a new all-time high (~$16B and climbing).
These are the top robotics companies/tickers
Robotics company investments just smashed a new all-time high (~$16B and climbing).
These are the top robotics companies/tickers
🛢WTI, Brent Crude Oil, NGAS Trading 📊
🚨🇺🇸🇮🇷 The U.S. just escorted a single tanker through the strait with a war's worth of aircraft overhead, and the math behind it tells you exactly what this fight is really about... Picture the package Washington put over one ship moving through the Omani…
This media is not supported in your browser
VIEW IN TELEGRAM
The Strait of Hormuz as soon as the stock market closes for the weekend
🇨🇭 Global One SI - Prompt
Bottleneck. The word of the moment. My whole feed is bottlenecks. $MU this. $SNDK that. $SIVE, $AAOI, etc. Fine. But why does the bottleneck exist in the first place? AI datacenters. No datacenters, no bottlenecks. And where does the supply actually…
Why doesn't $IREN need to rush a customer?
Easy.
IREN did exactly that once.
Horizon was built for $MSFT.
But now, IREN is a clean proxy of $NVDA.
Most neoclouds have to wait for a client before they build, because they need to tailor the datacenter to that client's spec.
No tenant, no blueprint, no shovel in the ground.
But from here on, it doesn't have to wait.
Because IREN is building to NVIDIA's DSX design.
It changes everything.
At GTC, Jensen unveiled NVIDIA Omniverse DSX.
It's the reference blueprint for building gigawatt-scale AI factories.
The building.
The power.
The cooling.
The networking.
All of it, co-designed with NVIDIA's infrastructure stack and captured as a single, repeatable recipe.
Read that word again.
Repeatable.
DSX was built so that operators can stand up AI factories from scratch, without reinventing the design every time, and know the result will run NVIDIA silicon at peak efficiency.
The thermodynamics, the power topology, the rack density, all solved in advance, validated in a digital twin before a single foundation is poured.
So what does that do for IREN?
It removes the need to wait for a customer to hand them a blueprint.
The blueprint already exists.
It's NVIDIA's.
IREN just builds to it.
And this isn't a side rumor. It's on NVIDIA's own product page: NVIDIA and IREN intend to deploy up to 5 gigawatts of DSX-aligned AI infrastructure across IREN's global pipeline over time.
Five gigawatts.
Pre-designed.
Pre-validated.
Customer-agnostic.
That's the unlock.
While other neoclouds sit idle waiting for a tenant to define their build, IREN can pour foundations now, to a spec NVIDIA already guarantees will be in demand.
They're not building a datacenter and hoping a customer shows up.
They're building the exact factory NVIDIA's entire ecosystem is standardizing around.
The customer is a detail.
The design is the moat.
Do not forget, NVIDIA owns the bottlenecks.
NVIDIA is the chokepoint.
And IREN is his son now.
He gonna get priority.
This is not financial advice. Do your own research.
I am long $IREN.
Easy.
IREN did exactly that once.
Horizon was built for $MSFT.
But now, IREN is a clean proxy of $NVDA.
Most neoclouds have to wait for a client before they build, because they need to tailor the datacenter to that client's spec.
No tenant, no blueprint, no shovel in the ground.
But from here on, it doesn't have to wait.
Because IREN is building to NVIDIA's DSX design.
It changes everything.
At GTC, Jensen unveiled NVIDIA Omniverse DSX.
It's the reference blueprint for building gigawatt-scale AI factories.
The building.
The power.
The cooling.
The networking.
All of it, co-designed with NVIDIA's infrastructure stack and captured as a single, repeatable recipe.
Read that word again.
Repeatable.
DSX was built so that operators can stand up AI factories from scratch, without reinventing the design every time, and know the result will run NVIDIA silicon at peak efficiency.
The thermodynamics, the power topology, the rack density, all solved in advance, validated in a digital twin before a single foundation is poured.
So what does that do for IREN?
It removes the need to wait for a customer to hand them a blueprint.
The blueprint already exists.
It's NVIDIA's.
IREN just builds to it.
And this isn't a side rumor. It's on NVIDIA's own product page: NVIDIA and IREN intend to deploy up to 5 gigawatts of DSX-aligned AI infrastructure across IREN's global pipeline over time.
Five gigawatts.
Pre-designed.
Pre-validated.
Customer-agnostic.
That's the unlock.
While other neoclouds sit idle waiting for a tenant to define their build, IREN can pour foundations now, to a spec NVIDIA already guarantees will be in demand.
They're not building a datacenter and hoping a customer shows up.
They're building the exact factory NVIDIA's entire ecosystem is standardizing around.
The customer is a detail.
The design is the moat.
Do not forget, NVIDIA owns the bottlenecks.
NVIDIA is the chokepoint.
And IREN is his son now.
He gonna get priority.
This is not financial advice. Do your own research.
I am long $IREN.
✍1
🇨🇭 Global One SI - Prompt
The robotics race won't be won by legs. Dancing, Kung Fu fighting robots capture mass attention. Hands unlock economic value. Most deployed robots today use grippers, which work in structured industrial environments. The real world isn't structured. Dexterous…
Who builds the robot hand?
A dexterous robot hand is a full hardware stack.
The visible part is the hand.
The hard part is inside: tactile sensing, motors, gearboxes, control boards, wiring, firmware and full-hand integration.
Key suppliers across the robot hand stack:
➤ 1. Tactile sensors
These give the hand contact feedback, pressure data and grip awareness.
• PaXini
• Hanwei
• Keli Sensing
• GelSight
• XELA
➤ 2. Actuation
These companies provide the motors and motion systems that move the fingers.
• Maxon
• Faulhaber
• MOONS’
• Leadshine
➤ 3. Transmission
This layer turns motor output into precise finger motion.
• Harmonic Drive
• Nabtesco
• LeaderDrive
• Laifui Harmonic
• Wuzhou Xinchun
➤ 4. Control
Control electronics manage finger motion, sensor input and low-level hand coordination.
• Zhen Ding
• Mektec
• Leadshine
• Avary Holding
• Dongshan Precision
➤ 5. Integrated hands
These companies build complete robotic hand systems.
SHARPA
• Linkerbot
• PaXini
• AgiLink
• Xynova
WUJI
• Shadow Robot
• Wonik Allegro
• DLR
• Inspire Robots
Robot hands are becoming one of the most important bottlenecks in humanoid robotics.
Walking gets attention.
Manipulation decides whether the robot can work.
A dexterous robot hand is a full hardware stack.
The visible part is the hand.
The hard part is inside: tactile sensing, motors, gearboxes, control boards, wiring, firmware and full-hand integration.
Key suppliers across the robot hand stack:
➤ 1. Tactile sensors
These give the hand contact feedback, pressure data and grip awareness.
• PaXini
• Hanwei
• Keli Sensing
• GelSight
• XELA
➤ 2. Actuation
These companies provide the motors and motion systems that move the fingers.
• Maxon
• Faulhaber
• MOONS’
• Leadshine
➤ 3. Transmission
This layer turns motor output into precise finger motion.
• Harmonic Drive
• Nabtesco
• LeaderDrive
• Laifui Harmonic
• Wuzhou Xinchun
➤ 4. Control
Control electronics manage finger motion, sensor input and low-level hand coordination.
• Zhen Ding
• Mektec
• Leadshine
• Avary Holding
• Dongshan Precision
➤ 5. Integrated hands
These companies build complete robotic hand systems.
SHARPA
• Linkerbot
• PaXini
• AgiLink
• Xynova
WUJI
• Shadow Robot
• Wonik Allegro
• DLR
• Inspire Robots
Robot hands are becoming one of the most important bottlenecks in humanoid robotics.
Walking gets attention.
Manipulation decides whether the robot can work.
🇨🇭 Global One SI - Prompt
Researchers at MIT, Stanford, and Nvidia built an AI agent that communicates telepathically. There's no words or language, just raw thoughts passed between agents. Right now, when two AI agents collaborate, they talk to each other like humans do. Agent…
Someone just rebuilt the entire Claude Code in pure Python and dumped it on GitHub for free.
→ Runs on any model (GPT, Gemini, DeepSeek, GLM)
→ 58.2% on SWE-bench Verified
→ 6x cheaper than the real thing
→ 180,000 lines. fully working
100% Open Source.
github.com/agentforce314/clawcodex
→ Runs on any model (GPT, Gemini, DeepSeek, GLM)
→ 58.2% on SWE-bench Verified
→ 6x cheaper than the real thing
→ 180,000 lines. fully working
100% Open Source.
github.com/agentforce314/clawcodex
✍1
Some sources are saying Fable 5 and GPT-5.6 may be cleared for public release as early as next week, including outside the US.
The whole point of the delay was probably to buy time against China.
That is one full month of lost distillation time for Chinese labs, while US AI labs are already using these models internally to train whatever comes next.
Brutal move if true.
The whole point of the delay was probably to buy time against China.
That is one full month of lost distillation time for Chinese labs, while US AI labs are already using these models internally to train whatever comes next.
Brutal move if true.
A crazy blog.
Chinese developers are buying Claude access through gray-market API transfer stations that can sell tokens at 5% to 10% of official prices while hiding the real user from Anthropic.
A transfer station is a middle server that takes a user’s prompt, sends it to Claude through overseas accounts, returns the answer, and collects payment through WeChat or Alipay.
The transfer station collects many Claude accounts through free credits, discounted accounts, shared subscriptions, overseas payment workarounds, fake verification, or sometimes stolen-card accounts.
It connects all those accounts behind one proxy, so Chinese users do not talk to Anthropic directly and only pay the proxy in RMB.
The cheap price comes from account farming, free-credit abuse, resale of unused quota, subscription splitting, possible stolen cards, and a darker trade where user prompts and outputs become training data.
So the price hugely cheap not because Anthropic is giving a discount; it is cheap because the transfer station lowers its own cost and creates extra hidden revenue.
The user thinks they are buying cheap inference, but the proxy may swap Opus for weaker models, inflate token use, or store private code, tool calls, reasoning traces, and business data.
The proxy may store user prompts, code, outputs, and tool traces, then sell or reuse that data for model training.
This breaks a core assumption behind KYC, account bans, and abuse monitoring: the AI company sees the proxy, not the real person, so banning one account leaves the upstream supply chain alive.
chinatalk.media/p/how-to-buy-cheap-claude-tokens-in
Chinese developers are buying Claude access through gray-market API transfer stations that can sell tokens at 5% to 10% of official prices while hiding the real user from Anthropic.
A transfer station is a middle server that takes a user’s prompt, sends it to Claude through overseas accounts, returns the answer, and collects payment through WeChat or Alipay.
The transfer station collects many Claude accounts through free credits, discounted accounts, shared subscriptions, overseas payment workarounds, fake verification, or sometimes stolen-card accounts.
It connects all those accounts behind one proxy, so Chinese users do not talk to Anthropic directly and only pay the proxy in RMB.
The cheap price comes from account farming, free-credit abuse, resale of unused quota, subscription splitting, possible stolen cards, and a darker trade where user prompts and outputs become training data.
So the price hugely cheap not because Anthropic is giving a discount; it is cheap because the transfer station lowers its own cost and creates extra hidden revenue.
The user thinks they are buying cheap inference, but the proxy may swap Opus for weaker models, inflate token use, or store private code, tool calls, reasoning traces, and business data.
The proxy may store user prompts, code, outputs, and tool traces, then sell or reuse that data for model training.
This breaks a core assumption behind KYC, account bans, and abuse monitoring: the AI company sees the proxy, not the real person, so banning one account leaves the upstream supply chain alive.
chinatalk.media/p/how-to-buy-cheap-claude-tokens-in
👀1
Austria is asking the EU to give Anthropic a European base because US AI controls have turned frontier-model access into a geopolitical dependency.
It is asking Brussels to explore a political and legal package that could make Anthropic run part of its business from inside the EU, with EU law, EU customers, EU capital, and EU-based infrastructure involved.
The hard part is that hosting Anthropic in Europe would not automatically move US ownership, export law, training infrastructure, or model governance outside Washington’s reach.
Austria gave no clear mechanism for how this would actually happen.
The catch is huge: even if Anthropic had servers and a subsidiary in Europe, US export-control law could still bind Anthropic because the parent company, core technology, employees, and model governance remain American.
Austria’s real claim is that Europe should not depend fully on a US company whose strongest models can be cut off by a US government order affecting foreign users.
It is asking Brussels to explore a political and legal package that could make Anthropic run part of its business from inside the EU, with EU law, EU customers, EU capital, and EU-based infrastructure involved.
The hard part is that hosting Anthropic in Europe would not automatically move US ownership, export law, training infrastructure, or model governance outside Washington’s reach.
Austria gave no clear mechanism for how this would actually happen.
The catch is huge: even if Anthropic had servers and a subsidiary in Europe, US export-control law could still bind Anthropic because the parent company, core technology, employees, and model governance remain American.
Austria’s real claim is that Europe should not depend fully on a US company whose strongest models can be cut off by a US government order affecting foreign users.
👀1
This media is not supported in your browser
VIEW IN TELEGRAM
Anthropic CEO Dario Amodei on Open-Source AI Models.
"I don't think open source works the same way in AI that it has worked in other areas. Primarily because with open source you can see the source code of the model. Here we can't see inside the model, it's often called open weights instead of open source to kind of distinguish that. But a lot of the benefits, which is that many people can work on it and that it's kind of additive, don't quite work in the same way.
So I've actually always seen it as a red herring. When I see a new model come out I don't care whether it's open source or not. If we talk about Deep Seek I don't think it mattered that Deep Seek is open source. I think I ask, is it a good model? Is it better than us at the things that matter? That's the only thing that I care about.
It actually doesn't matter either way. Because ultimately you have to host it on the cloud. The people who host it on the cloud do inference. These are big models, they're hard to do inference on.
When I think about competition I think about which models are good at the tasks that we do. I think open source is actually a red herring.
It's not free. You have to run it on inference and someone has to make it fast on inference."
"I don't think open source works the same way in AI that it has worked in other areas. Primarily because with open source you can see the source code of the model. Here we can't see inside the model, it's often called open weights instead of open source to kind of distinguish that. But a lot of the benefits, which is that many people can work on it and that it's kind of additive, don't quite work in the same way.
So I've actually always seen it as a red herring. When I see a new model come out I don't care whether it's open source or not. If we talk about Deep Seek I don't think it mattered that Deep Seek is open source. I think I ask, is it a good model? Is it better than us at the things that matter? That's the only thing that I care about.
It actually doesn't matter either way. Because ultimately you have to host it on the cloud. The people who host it on the cloud do inference. These are big models, they're hard to do inference on.
When I think about competition I think about which models are good at the tasks that we do. I think open source is actually a red herring.
It's not free. You have to run it on inference and someone has to make it fast on inference."
✍1