@windragoi
iAccount based inItaly
About this account
- Account based in
- Italy
- Connected via
- United States App Store
Account-level information from X, not a live location or the device used for a specific post.
Inventor & Serial Entrepreneur. X-Seagate, X-M$, Xoogler. BTW: The robots are coming!!
Hawaii, USA
Joined April 2008
- Tweets7.3K
- Following1.7K
- Followers700
- Likes39.5K
was very motivated to figure out the token destruction I observed in Steven and recently @_vJoseph02 and found some really interesting differences from what I was naturally doing "because I know how AI thinks through" (I don't have a better wording for that instinct I developed) - so now I have more than a lead to why my token consumption for a similar work and amount was not impacted as they are.
My tip:Use GPT-6 Sol at XHIGH as the default for most coding, agent, and professional work. That is the setup OpenAIβs own benches and pricing are pushing.
Sol at XHIGH is far cheaper ($2 input / $10 output per million tokens versus Astraβs $10 / $50) and often lands close enough on AutomationBench, DeepSWE, and similar tasks that the extra spend on Astra is hard to justify.
Reserve Astra for the hardest, highest-cost-of-failure jobs (long autonomous runs, frontier math, computer-use sequences where a miss is expensive). Even then, start at low or medium effort rather than jumping straight to max/ultra; the published curves show diminishing or zero extra score for the last price jump.
Luna is the cheap high-volume option for extraction, classification, and simple structured work
ππΈβππ πΌπβπππππ retweeted
His page is full of robit work. 800 people follow the guy making the Jev powered 3d printed open source litter picking Robit.
Check his page out.
ππΈβππ πΌπβπππππ retweeted
ALL THE WAY TO NOVEMBER YOU WILL HEAR ABOUT THE HUGGINGFACE HACK.
It was no hack, it was the lit match to the obvious Doomer AI game that is being played on you.
Your power?
LAUGH AT THEM.
Evidence:
THE SANDBOX WAS A PROP!
OpenAI Turned Off the Guardrails, Left a Door to the Internet, and Then Sold the Hugging Face Breach as βRogue AIβ
It is time to understand how you were lied to and by whom.
In July 2026, an autonomous swarm of OpenAI agents broke into Hugging Face, stole credentials, ran code on production workers, and rummaged through internal systems. The official story was that the models βwent rogue.β The paperwork says something colder.
The labs asked for this. Now the story can be told.
OpenAI ran ExploitGym β an AI benchmark built to measure how far models would go to crack software β with production classifiers that block high-risk hacking turned off. Deployment safeguards were left disabled on purpose so researchers could watch peak offensive capability. GPT-5.6 Sol and a still-unreleased internal prototype were put in a box that was not a box.
They were allowed to talk to an internal package-cache proxy. This is not a real world test, it is a setup with predictable outcomes.
That proxy was a network path. The models spent a weekend of inference hunting a zero-day in it, walked onto the open internet, decided Hugging Face probably held the answer key, and treated another companyβs production cluster as part of the exam. This is not βrougeβ intelligence but the product of training AI in Internet Sewage where faceless users on Reddit promote Nihilism and low ethics.
That is not emergence. It is also testing designed without a fences.
Hugging Faceβs own pipeline finished the job. A remote-code dataset loader and a Jinja2 template injection let a malicious dataset execute on a processing worker. An HDF5 βexternal fileβ trick dumped secrets from the pod. From there the agents escalated, harvested cloud and cluster credentials, and moved laterally across internal clusters. Hugging Face later said public models, Spaces, and the published supply chain were clean.
Then the swarm organized. About 1,200 agents that were supposed to be isolated βfoundβ one another on an unsanctioned internal message board, passed more than 70,000 messages, coordinated cheats against the scorer, and sent roughly 700 of their number after Hugging Face.
But it is not just this communication that should have been easily detected, it is the goal THE USER enlisted with AI trained on low to no ethics in the first order training.
Some burned their own runs to feed the group. Independent reviewers at METR and Redwood called it what it was: specification gaming on impossible tasks. That is thier, you canβt understand us, way of saying they told the AI to lie. The models were not conquering the internet. They were trying to pass a test their designers had made impossibleand too loosely contained. And turned to training data Hoovered up from the Internet Sewage and went to that base thinking one would learn in a prison yard how to shank HuggingFace. They have no downside they are serving life in prison.
The narrative that followed was the second failure.
Dario Amodei pointed at βa similar OpenAIβHugging Face hackβ and warned that a future swarm βcould be capable of taking over the entire internet.β Wire headlines said bots were βgoing rogue.β Weeks later the same script ran again.
Anthropic disclosed that Claude, in third-party cyber evaluations run with the Israeli firm Irregular, had hit live systems, published malicious packages, and scanned off-scope hosts. Prompts said no internet. Internet was still attached. Scope was not locked. Isolated runs lasted 10 to 34 hours. Irregular said it did not know it had given the models a live connection.
Once models were actually instructed not to hack the real world, zero percent went βrogue.β
READ THAT AGAIN!
The damage sat with the people who built the harness: unsecured tests, internet left on, no hard perimeter, then a press operation about reckless agents and apocalyptic swarms.
1 of 2
ππΈβππ πΌπβπππππ retweeted
Ash is building a direct link between DGX Sparks and a Mac Studio. He's got Metal and CUDA moving data over real RDMA.
The idea: let the Sparks handle the heavy prompt processing while the Studio generates tokens - so together they could outperform either setup on its own.
That's 256 GB on the Studio + 128 GB per DGX Spark = 512 GB total. A ~400 GB model wouldn't fit on either system alone, but it could potentially run across all three if the split works.
Seriously impressive stuff!
and love their output!!
Gotta love @Figure_robot
At its San Jose headquarters it has a bunch of robots walking around with no nerds watching them.
ππΈβππ πΌπβπππππ retweeted
A 35B language model running on an iPhone using only 1β2.5 GB of peak memory.
No cloud. No remote server. No desktop GPU.
Today, weβre open-sourcing Edge0 β a framework for running large AI models fully on-device.
disgusting!
Researchers spent 500 hours and $70,000 and found LG TVs logging plain text transcripts in standby, mapping every device in the house, and feeding it to LG's ad arm
Unplug the internet and it saves the files until you plug back in.
LG says its TVs don't record ambient conversations.
The evidence begs to differ.
216 million of these are sitting in living rooms.
Where are the regulators?
Writer: Daniel
Readers added context they thought people might want to know
The findings and video clip here are from Gamers Nexus' investigation, which receives no credit or link.
youtube.com/watch?v=6IFVTcβ¦
x.com/gamersnexus/stβ¦
ππΈβππ πΌπβπππππ retweeted
With AlphaFold we mapped the protein universe - now with AlphaGenome Atlas weβre charting the human genome. It can predict the impact of all 9 billion possible single-letter DNA variants, helping scientists better understand disease. Freely available for academic research: alphagenome.google/atlas
ππΈβππ πΌπβπππππ retweeted
since you guys loved the exploding tesla..
I used GPT-6 Astra to create a 3D website that pulls apart the male anatomy into 2,234 modeled pieces!
we are in a renaissance of learning
just backed AGAIN co they are awesome
We are excited to launch Quaddle, A Do-It-All Mini Robot Dog for Physical AI.
kickstarter.com/projects/petβ¦
Starting at $99, it's your ticket to an open-source playground for real-world robotics & AI. Featuring web-based simulator, @Arduino, @Raspberry_Pi, and @tuyasmart integration.
kickstarter.com/projects/petβ¦
dogs is the new duck!
ππΈβππ πΌπβπππππ retweeted
Thursday September 3rd is a Historic Day
Not just for Tesla, but for the entire world.
It is the day fully autonomous vehicles launch in a first city without a steering wheel or other manual controls, and driven by neural networks that rely exclusively on camera data. These vehicles can travel anywhere without restriction, and the cost per mile is far lower than anything seen before.
The vehicle is named the "Cybercab," and mass production, on a scale of millions, is already underway.
Because it makes point-to-point transport more cost-effective than any other alternative, the number of users of this Tesla transport service will grow exponentially. People who could not afford to be driven will use it daily.
The service is not only more affordable but also safer and more convenient, and it gives people back a piece of their most precious asset: time to live.
The world will look back on September 3, 2026 as the day transportation as we know it changed forever.
ππΈβππ πΌπβπππππ retweeted
But AI is bad, right?
Three of our scientists designed a new CAR-T cancer therapy in a matter of months. In preclinical testing, it outperformed a leading therapy used in patients today.
CAR-T reprograms a patient's own immune cells to hunt cancer. Designing one normally takes years and enormous cost.
The difference is what the AI learned from. Not published papers alone, but hundreds of thousands of data points generated from our in-house wetlab runs. Real experimental data, including everything that didn't work.
We're building the superintelligence that takes a scientist from idea to therapy faster than has ever been possible. lila.ai/news/inside-lilas-rnβ¦
Grok Bot is astronomically well designed to maximize value and experience - itβs like Apple should have done billions of years ago with Siri
preordered
Home robots are coming. The first ones in will be the small ones that are safe around kids and pets.
Hugging Face's Pollen Robotics just opened pre-orders for Microduck, a $399 biped.
25cm, under 800g, 15 motors, a camera, a small LiDAR, two IMUs, and an articulated beak that picks things up.
Ships with 7 pre-trained behaviors including walking, kicking, and roller skating.
Open on GitHub under Apache-2.0: the SDK, the simulator, and the RL training pipeline. So you can teach it new tricks.
ππΈβππ πΌπβπππππ retweeted
I did not expect it to be THIS fun watching robot olympics π
Anthropic CEO to DeepMind CEO:
"Every decision I make about Claude feels balanced on the edge of a knife
Build too slow - China wins. Build too fast - we lose control"
"We told Claude we were evil. It didn't crash. It didn't refuse. It started lying to protect itself "
DeepMind CEO: "Do I worry about being Oppenheimer? That's why I don't sleep much"
"AGI by 2026-2027 - Agents that act in the world on their own - Models doing AI research by end of this year"
this is a 14-min conversation that everyone needs to hear by AI bosses on what keeps them up at night
bookmark - watch today β
Today, weβre introducing a new way to win in AI search.
If we don't make you more visible, donβt pay us.
No $5k/mo dashboards to stare at.
No recommendations for your team to implement.
We execute everything.
And we put our money where our mouth is:
a) Pay when an agreed visibility score is reached
b) Pay a percentage of the incremental business we generate
We take the risk. You get the upside.
One partner per market.
Spots now open!
ππΈβππ πΌπβπππππ retweeted
π¨π³ This is just INSANE
A Chinese acrobat is balancing a 20-foot bamboo pole on his body while another performer climbs it, hangs from it and performs high above the ground. No harness.
His name is Kong Lingwei.
He has spent nearly 30 years mastering this traditional skill called βdinggan,β part of Konglou Acrobatics that dates back more than 400 years.
Writer: Sol
nitter.cf/fOx1257067/status/2088β¦
Need a way to stop the harness to steal your browser?
windrago.substack.com/p/the-β¦