Nothing Works the Way You Think

A study of 25,114 manuscripts submitted to 49 BMJ biomedical journals found that only about 5.7% of authors admit to using AI tools like ChatGPT. Most use it to clean up their writing, but the researchers suspect the actual usage is much higher. We need better ways to detect when a machine helped write a scientific discovery. (More)

Scientists are sounding the alarm: AI models that work well in controlled settings often fail in the real world. A camera-trap wildlife model that was 97% accurate at identifying animals in its training set might drop to 36% accuracy in a different forest. It turns out the natural world is much more complex than our testing data, leading to misleading claims. (More)

AI chatbots can be creative partly because they sometimes get things wrong. A study found that some of the techniques we use to stop AI from lying can suppress its ability to brainstorm original ideas , though other techniques had the opposite effect, actually boosting creativity. Scientists need to find a balance where an AI is factually accurate but still divergent enough to help with breakthroughs. (More)

While humans win on video, machines are almost impossible to beat when it comes to static images. Convolutional Neural Networks (CNNs) achieved a staggering 97% accuracy in detecting fake faces in photos, while humans were essentially just guessing at chance level. It's a reminder that machines see pixel-level artifacts that are invisible to our eyes. (More)

Bird watchers have been confused by the Cercomacra bird complex for years because they all look identical. But by using AI-powered song analysis (BirdNET) and plumage mapping, scientists finally identified two brand-new species that were hiding in plain sight. It's a perfect example of how tech is helping us map the true diversity of life. (More)

The SETI Institute thinks alien radio signals might be getting blurred by space weather before they reach us. Turbulent plasma near distant stars can spread out a sharp radio tone into a faint, wide signal that our current searches would totally miss. We're now redesigning our searches to look for these smeared signals. (More)

Japan is doubling down on international science by building globally visible research centers through the WPI Initiative. These hubs are designed to be science magnets, attracting the world's top talent to Japan to collaborate on everything from robotics to material science. (More)

Artists can now use a tool called MMMC to cloak their singing voices. It adds tiny, invisible modifications to audio that make it impossible for an AI to produce a high-quality clone. In a world where celebrity voices are being used without permission, this tool gives creators a way to defend their digital identity. (More)

Your brain starts as one cell and grows into 170 billion. Scientists used to think it was all about chemical signals, but a new theory suggests it's more like how families settle in a new country. Cells know where to go simply because they stay near their relatives, creating a massive structure without anyone ever being the boss. (More)

This is wild: if you tell someone their drink has sugar in it, they actually enjoy artificial sweeteners more. A neuroscience study showed that your brain's reward centers light up based on what you expect to taste, not just what's actually there. It turns out mind over matter works even for your taste buds. (More)

MIT engineers figured out that they can steal your tacit knowledge (the skills you have but can't explain) just by watching your eyes. When experts look at a complex image, they focus on the most important parts without realizing it. By showing those eye maps to novices, the researchers turned beginners into experts almost instantly. (More)

Most AI is built on simple running averages, but human brains are much cooler than that. New research shows we create high-dimensional memory variables that track our entire history of wins and losses. This helps us adapt to a changing world in ways that simple algorithms just can't match. (More)

AI helps you the most if you actually know what you're bad at. A study found that well-calibrated workers (those who understand their own limitations) saw a 10% jump in performance when using AI. Highly skilled but arrogant workers actually benefited the least, proving that AI rewards those who know when to ask it for help. (More)

The explosion of AI art can either help or hurt society, depending on the quality gap. If the tech helps beginners get better, it's a win for everyone. But if the market gets flooded with mediocre content that crowds out the pros, everyone loses. We need drastic tech improvements to make this a net positive. (More)

Deep inside a buried part of your brain called the insula, Mayo Clinic researchers found a secret movement map. They discovered specific spots linked to your hands, feet, and tongue. This hidden map is going to be a game-changer for surgeons trying to avoid damage during procedures. (More)

For a long time, we thought that if you were good at directions, parts of your brain would actually be bigger. But a new study using AI just proved that for young adults, there's no connection at all between brain structure and navigation skills. Basically, having a big brain doesn't mean you won't get lost in a parking lot. (More)

Thank you for reading!

Stay curious and in the loop.