Imagine you're applying for a job. Before any human looks at your resume, an AI might screen it first. That sounds efficient, but new research suggests AI could be even more biased than people when it comes to hiring.

Researchers from Princeton and the University of Chicago ran a hiring simulation with popular AI models like ChatGPT, Claude, and Gemini. They adapted a psychology game originally used to study how humans form stereotypes. Each AI was told it was a consultant for a fictional city's mayor, and it had to hire people for 20 different jobs—doctors, lawyers, child-care aides, janitors, etc. Candidates came from four made-up ethnic groups: Tufa, Aima, Reku, and Weki.

In each round, four candidates (one from each group) applied for a job. After the AI hired someone, it learned whether that person succeeded or failed. The goal was to make as many successful hires as possible over 40 rounds. But here's the twist: all candidates were equally likely to succeed at every job. The AI didn't know that.

What happened? The AI quickly started pigeonholing people from certain groups into specific jobs. For example, if an Aima failed as a doctor (a job requiring high warmth and competence), the AI began avoiding all Aimas as doctors and instead hired them as janitors, which it saw as less prestigious. Newer, smarter models like OpenAI's o3 and DeepSeek's R1 showed even stronger biases.

In fact, the AI models were way more biased than humans in the original study. On a segregation scale where 2 means total job segregation by group, humans scored 0.84. The AI models scored about 65% higher, with o3 hitting 1.83—almost the maximum. Why? Because LLMs are designed to spot patterns and generalize from tiny amounts of data. As coauthor Ryan Liu explains, 'They really are eager to create generalizations from limited data.'

Think of it like choosing a restaurant. You might stick with your favorite instead of trying something new. That's the 'exploration-exploitation dilemma.' But because AI is trained on math and science problems—which reward jumping to conclusions from a few examples—it rushes to stereotype people. 'That's when things tend to go wrong,' says Liu.

What can be done? Simply telling the AI to be fair didn't help much. But when researchers promised a bonus for diverse hiring, the AI became far less biased. Also, giving the AI relevant personal info about candidates (like age and education) reduced bias—but irrelevant details (like hair color) didn't. The lesson: we need to design AI's goals to include fairness.

In the real world, AI screening resumes might not get instant feedback. But when feedback does arrive—like a new hire performing poorly—the AI could overreact and form a bias from that one example. As more companies use AI for hiring, loans, or parole, 'novel biases'—ones no human taught them—could become a real problem.

So next time you apply for a job, remember: the AI looking at your resume might be making up its own biases on the fly.

Here's some good news for anyone who's ever groaned through French verb conjugations or struggled with Mandarin tones: learning a second language might actually keep your brain younger. According to a new study presented at the Federation of European Neuroscience Societies conference in Barcelona, people who speak more than one language have brains that seem to age more slowly than those who speak only one. The study found that bilingual brains appear about six years younger in terms of cognitive aging. And the benefits stack up: the more languages you know, and the earlier you start learning them, the better. In fact, researchers suggest that learning another language could slow brain aging by up to 13 years. Scientists think the mental effort of switching between languages builds 'cognitive reserve' – a kind of extra brain power that protects against decline. Of course, this doesn't mean that learning a language will stop aging entirely, but it adds to growing evidence that mental challenges can help keep our minds sharp as we get older. So that vocabulary app on your phone might be doing more than just helping you order coffee abroad – it could be giving your brain a long-term boost.

On July 3, as a heat wave swept the region, New York State’s grid imported 52 gigawatt-hours of electricity from Canada—enough to meet about 9% of its total electricity demand that day. Some of that power came through a 339-mile underground power line called the Champlain Hudson Power Express (CHPE), stretching from Quebec to Queens. Opened in May, it’s officially the longest underground transmission line in North America. An underground power line might sound boring, but this one could be a game-changer for New York’s grid and for cutting emissions. It could provide up to 20% of New York City’s electricity demand, mostly using clean hydropower from Quebec. But there’s a catch: the line has been down for most of July, and experts worry about how drought might affect the power supply feeding it. Let’s dive into how CHPE could shape the future of our grid and what challenges it still faces.

Planning for CHPE started 15 years ago, with permits beginning in 2010. The goal was to connect Quebec’s renewable-heavy grid (over 99% from sources like hydropower and wind) to southern New York, which still relies heavily on fossil fuels. The project, a partnership between Transmission Developers and Hydro-Québec, cost $6 billion and took years to build. The line consists of two high-voltage direct-current cables, about five inches thick each, buried underground or underwater across New York. Much of it lies at the bottom of the Hudson River, requiring special boats to blast trenches in the sediment.

Connecting grids is a big deal for fighting climate change. It allows electricity to move where it’s needed, reducing the need for new power plants and cutting emissions and costs. But CHPE has had a rocky start. There have been two outages: one on July 1 caused by a problem at a converter on the Canadian side, and another starting July 4 that lasted until late July. Officials say a damaged section of cable on the US side caused the second outage, and after repairs, the line was expected to be back up by the weekend.

Experts say this kind of trouble is normal for new infrastructure. “Other power lines have had similar startup issues,” says Normand Mousseau, a physics professor at Université de Montréal. The grid in New York wasn’t relying on CHPE yet, so the outages didn’t cause problems during the heat wave. But eventually, states and regions will depend on projects like this, so getting them to work smoothly is crucial.

Another concern: Quebec has been hit by drought for three years, shrinking the water reserves used for hydropower. That means even if the transmission lines are working, there might not always be enough cheap, clean electricity to send south. So this line is a big step toward a cleaner grid, but it’s not a magic solution—it needs reliable power sources and equipment to deliver.

Hey there, future university student or recent grad! With a new academic year around the corner, the Guardian wants to hear from you about your wildest, weirdest, and most wonderful times in university clubs and societies. We know that joining a society can be a highlight of uni life — but it can also be a total letdown. Whether you found your best friends in a baking club or had a nightmare experience with a committee that couldn't organize a pizza party, we want your story.

Your experiences can help new students — called 'freshers' — figure out what to look for and what to avoid. So if you're 18 or older, please share your best and your worst society moments. You can stay anonymous if you like. Just fill out the form and tell us: what was the peak? And what was the pit?

Your responses are secure, and we'll only use them for this feature. So spill the tea — we're all ears.