Hacking Science

Canadian researchers at the University of Waterloo have developed a tiny, flexible robot, about the size of a grain of rice, designed to help with certain kidney stones. The soft robot contains micromagnets and a natural enzyme, and is guided into place using a magnetic arm from outside the body. Once positioned, it works to gently dissolve uric-acid stones over a few days, allowing them to pass naturally.
So far, the technology has been tested in a life-size 3D-printed model, not yet in patients, so more research is needed before it becomes widely available. The findings were published in the journal Advanced Healthcare Materials.
Sources: University of Waterloo News, Advanced Healthcare Materials

Chinese electric vehicle giant BYD has officially patented a new, artificial intelligence-powered safety technology designed to detect people, animals, and other living organisms hiding underneath parked vehicles before they begin moving.
Filed under application CN122200729A and published by China’s National Intellectual Property Administration on June 12, 2026, the technology aims to completely prevent accidents involving children or small pets sheltering under cars.
The system utilizes underbody cameras to create a two-stage computer vision process.
When the vehicle is turned off, the system captures a baseline reference image of the car's underside, including fixed components like the battery pack, suspension, and chassis.
When the vehicle is started up again, real-time cameras scan the underbody and compare it to the original reference photo. Instead of wasting processing power on the entire underbelly, deep learning algorithms isolate only the specific regions showing visual changes.
The AI then focuses exclusively on those altered zones to analyze the characteristics of the object.
It is specifically trained to filter out harmless environmental variables like shadows, dust, leaves, and mud, drastically reducing the risk of false alarms.
By recognizing movement patterns, body shapes, and behavioral cues, the technology can accurately identify a trapped or resting living organism and instantly issue an alert to the driver before the wheels turn.

A robot guided by an AI just performed surgery with near-perfect accuracy and no human intervention.

In a major step toward autonomous surgical systems, researchers at Johns Hopkins University have successfully trained an AI-guided robot to perform gall bladder removal on a dead pig with near-complete autonomy.

The dual-layer AI model—trained on 17 hours of human surgery footage—translated video input into verbal commands like “clip the second duct,” which were then converted into precise tool movements.

The robot completed all 17 surgical tasks across eight procedures with 100% task success, demonstrating real-time self-correction and only limited human intervention.

While full autonomy in live human surgeries remains years away, the milestone marks a significant advance in surgical robotics.

Experts emphasize that while the robot occasionally needed tool changes or adjusted missed grabs, its ability to detect and fix its own mistakes shows the growing reliability of AI-assisted procedures. With regulators expected to weigh in heavily before human applications, researchers plan to move next toward testing in live animals. The long-term goal? Safer, more precise surgeries that reduce complication rates and ease strain on surgical teams.
source

Journal reference: Ji Woong (Brian) Kim. Science Robotics. 10 (4). SRT-H: A hierarchical framework for autonomous surgery via language-conditioned imitation learning

A new space startup plans to sell on-demand sunlight at any hour, beamed directly from Earth's orbit.
California-based aerospace startup Reflect Orbital is looking to revolutionize how we experience the night by launching a constellation of satellites equipped with ultra-thin reflective mirrors.
Founded by former SpaceX engineer Ben Nowack, the company aims to make sunlight "programmable," allowing users to request a localized beam of light simply by entering GPS coordinates on a website.
The system is designed to redirect sunlight down to target areas roughly five kilometers wide, offering brightness ranging from a gentle full moon to peak daylight. Initial applications focus on expanding solar power generation into the evening hours, assisting disaster response teams, and lighting remote industrial worksites.
While the startup has generated massive commercial interest—receiving more than 260,000 inquiries and securing millions in venture capital—the concept of "sunlight-as-a-service" has sparked intense debate.
Environmental scientists and astronomers warn that artificial orbital illumination could severely disrupt nocturnal ecosystems, disorient migrating birds, and damage human circadian rhythms.
Critics argue that altering natural cycles of darkness constitutes a new, unregulated form of light pollution. Despite these concerns, Reflect Orbital is moving forward with plans to deploy its first test satellites, bringing humanity one step closer to a future where darkness is optional.
source: McKie, R. (2026). Satellite mirror plans could disrupt sleep and ecosystems worldwide, scientists say. The Guardian

AI generated image of a cow, 12 years of progress

A teenager in the United States started publishing software at 14 in 1998, built the entire online infrastructure for the Occupy Wall Street movement in 2011, joined Google as a software engineer, quit in 2018, and then spent five years writing a C library that does something the entire industry said was impossible.
Then she combined it with llama.cpp and shipped the easiest way on the planet to run a large language model on any computer.
Her name is Justine Tunney.
Here is the story, because almost nobody outside the low level systems world knows what one engineer has built.
Justine was born in 1984. She started writing and publishing software at 14, back when distribution meant uploading binaries to BBS systems and chat networks, She picked up the handle jart, which she still uses on GitHub today. She did the work most teenagers her age were not doing, She read the systems programming literature, She studied compilers She fell in love with C.
In July 2011 she registered the occupywallst dot org domain. Within weeks the protest movement that began in Zuccotti Park in New York had become a global phenomenon, and her infrastructure was the digital backbone of the entire thing, She handled the social media, the website, the donations, the coordination. She built the platform that pushed the movement to reach millions.
After Occupy she joined Google as a software engineer, She worked on TensorBoard, the visualization tool for TensorFlow, and on site reliability for Google infrastructure,
She stayed for years,
Then in 2018 she left Google Brain to work on a personal project.
The project was called Cosmopolitan Libc.
Cosmopolitan does something most C programmers would tell you is mathematically impossible, It lets you compile a C program once and have the resulting binary run natively on Linux, Windows, macOS, FreeBSD, OpenBSD, and NetBSD with no modification. One file, Six operating systems, No virtual machines, No interpreters, No recompilation. The technique she invented is called Actually Portable Executable.
The implications are wild, Cosmopolitan binaries violate every assumption about how operating systems load programs, They are at once a Windows PE file, a Linux ELF binary, a macOS Mach-O binary, and a shell script.
The same bytes run on every platform.
For five years she worked on it mostly alone. She funded the development partly through Mozilla's MIECO program, which sponsored her work on Cosmopolitan 3.0, released on October 31, 2023.
A month later she shipped llamafile.
llamafile is what happens when you combine Cosmopolitan with llama.cpp. You take any LLM weights file in the standard GGUF format, you wrap it in Justine's binary, and you get a single file that runs on six operating systems without installation. No Python. No CUDA setup. No dependency hell. Just one file that you double click and it works.
Mozilla launched it as an official project of their innovation group on November 29, 2023. It went viral immediately.
The repository, hosted at github .com/mozilla-ai/llamafile, now has 24,600 stars. The license is Apache 2.0.
Justine kept shipping. She added GPU support to Cosmopolitan, a task systems engineers thought would require rewriting the whole thing. She added dlopen support, another thing nobody else had figured out.
She wrote whisperfile, a single file version of OpenAI's Whisper speech-to-text model based on the same architecture.
Her GitHub profile lists projects most engineers would consider impossible. sectorlisp, a Lisp interpreter that fits in a boot sector. blink, the tiniest x86-64-linux emulator on Earth. bestline, a teletypewriter command session library. redbean, a complete web server inside a single zip file.
A teenager who shipped software in 1998 grew up to write the C library that the entire local AI movement now runs on top of.
She did most of it alone, and most peoples in Ai don’t know her name.

China-based developers have introduced Photon Matrix, a LiDAR-guided mosquito defense system capable of detecting and neutralizing up to 30 mosquitoes per second. The device uses Light Detection and Ranging (LiDAR) sensors to scan the environment, identify flying insects, and calculate their size, position, speed, and trajectory within approximately three milliseconds.

Once a mosquito is confirmed as a target, a precision laser is directed toward it for elimination. The system is programmed to avoid larger objects such as humans and pets and is available in versions covering up to 3–6 meters. Developers say the technology could support future mosquito-control and disease-prevention efforts.

More than 10 billion devices run on his idea.
He made $0 from every single one and he planned it that way.
Meet Ajay Bhatt,
- Indian-American engineer. Born 1957 in Vadodara, Gujarat.
- Came to the US with a master's degree and joined Intel in 1990.
- One frustrating night, he couldn't connect a printer for his daughter's homework.
- He asked: why isn't there ONE universal port?
- His boss said it would never work. Told him to drop it.
- He didn't.
- Built it with fellow Intel engineer Bala Cadambi.
- Then united 7 fierce rivals Intel, Microsoft, IBM, Compaq, DEC, NEC, Nortel behind one shared standard.
- Apple fought it with FireWire. USB was cheaper. USB won.
- USB 1.0 launched in 1996. He went on to build USB 2.0 and 3.0.
- Intel made it royalty-free free for the entire planet.
- Bhatt earned not a dime in personal royalties. By choice.
- 2009: Intel made him a rock star in a viral ad played by a hired actor, not him.
- 2025: India finally honored him with the Padma Shri.
The man who connected the world.
And asked for nothing in return.

PewDiePie trained a frontier model at home
Now he released Odysseus, a self-hosted AI workspace

- be PewDiePie
- play games on YouTube and scream 24/7
- become a meme reviewer
- get unfathomably famous
- f**k that, a family guy now
- retire
- move to Japan with beautiful wife
- mfw, a dad now
- as a dad you must do dad things
- scratch that
- "as a dad must do frontier AI research"
- goal is to beat GPT-4o at coding (16% on Aider)
- buys $30000 GPU setup
- reads DeepSeek paper
- decides to start massive GitHub scraping and data augmenting run
- not good enough
- read Magicoder paper
- generate tons of synthetic coding data
- train a new model
- guuuuuh. the data made the model worse
- mfw just wasted months for nothing
- decides to lock in and try again
- makes model worse again ffs
- try again
- finally beating GPT-4o on data (16.1%) few months ago
- not satisfied
- should simply train a reasoning model
- reads more papers
- start experimenting with more synthetic data
- Mhhh something doesn't smell right
- house almost burned down due to power connector
- shrug
- just buy a new one
- mfw computer is now crashing 24/7 generating synthetic data
- new plan: just call DeepSeek API for high quality synthetic data
- train model again
- 17.2%
- performance fluctuates slightly on each eval run
- big brain idea: repeat eval until randomly reach 18%
- s**e actually got 19.6%
- feelsgoodman.png
- nvm the benchmark was contaminated and was training the wrong base model the whole time
- rerun everything again
- new score: 4.4%
- you read that right REEEEEEEE
- almost get a heart attack
- have you tried plugging the device off and back on again?
- change nothing and just retrain again
- 25.3 %
- LETS FU***NG GOOO
- realize that 1/3rd of the benchmark was not running. guuuh
- scared sh*tless it would score below 10% again
- run yet again. the whole thing this time
- Thirty f**king six percent
- accidentally beat Gemini 2.0 Pro Exp and GPT-4.1 mini
- "want moaaaar"
- finds some more post-training data
- 39% babyyyy
- realize at the end that was just benchmaxxing Aider polyglot
- next quest: run SWE-Bench and other coding benchmarks
- " failed a thousand times, but prevailed in the end"
- just a little sad side-quest
- runs huge models locally like Qwen 235B and GPT-OSS 120B
- builds custom web UI because existing ones aren’t good enough
calls it Odysseus
- open sources it on GitHub Odysseus
- even has an AI council where models debate and vote on the best answer
- runs everything offline
- instantly gets thousands of stars

A MIT student told me he can read a 400-page textbook in two hours and retain more than people who read every word.

He doesn't skim.

He interrogates the book like a witness by using a free tool called NotebookLM.

I didn't believe him either. Then he showed me exactly what that means.

Most people open a textbook on page one and march forward. Chapter by chapter. Page by page. They finish exhausted, retain maybe 20% of it, and feel like the problem is their focus or their memory.

It's not. The problem is the approach.

A witness doesn't volunteer everything they know. You have to drag it out of them with the right questions. The same is true of a textbook.

Here's his exact process.

He uploads the entire textbook into NotebookLM before touching a single page. Then he runs one prompt before reading a word.

"What is the central argument this book is making, and what would a reader need to believe for that argument to fall apart?"

That question does something most students never do before opening chapter one. It hands him the skeleton of the entire book in under two minutes. He knows where it's going, what it's trying to prove, and exactly where to look for the cracks.

Now he's not a reader. He's a cross-examiner.

He goes into the book with three targeted questions NotebookLM helped him generate.

"What do I already believe about this subject that this book directly challenges?"
"Which chapters contain the evidence the author relies on most heavily?"
"Where is the author's argument weakest?"

He reads only the sections that answer those questions. Everything else he skips without guilt.

Then comes the move that makes it genuinely unfair.

Every time he finishes a section that matters, he closes the book and runs one more prompt.

"Based on what I just read, what question would expose a student who understood the surface of this argument but missed the logic underneath it?"

He answers that question from memory before moving on. If he can't answer it, he goes back. If he can, the concept is locked.

He told me most students highlight because it feels like learning. NotebookLM helped him realize highlighting is just decoration on top of forgetting. The only thing that actually works is being forced to reconstruct what you read without looking at it.

By the end of two hours he hasn't read every word. He's built a complete map of the book's argument, identified exactly where his prior beliefs were wrong, and retained every section that mattered because he was forced to reconstruct it himself.

The students who read every word finished the same book.

But they were passengers.
He was driving.

There's a difference between covering material and owning it.

Most students spend four years covering it.