NEWS YOU CAN’T MISS

Can AI Really Predict Your Next Decision?

Researchers in Germany have built an AI model designed to make decisions like a human.

Most AI systems are trained for specific jobs - like writing text, generating images, or formulating research. They can be adapted to new tasks, but don’t usually perform well outside their original purpose.

Centaur takes a different approach; it was built to mimic how people think across a variety of situations, including how we weigh risks, remember things, or deal with uncertainty.

To create it, the team fine-tuned a large language model using data from 160 psychology studies - more than 10 million decisions made by over 60,000 people across different tasks.

The result is an AI that simulates human thought. In early tests, Centaur could predict people’s choices even in tasks it hadn’t seen before, something most models struggle with. It also outperformed Llama and 14 other leading psychology models in most instances.

Why it matters: In areas where it's difficult to study real people (like young children or those with mental health conditions). Centaur could offer a safe, low-risk way to explore how people might behave. Researchers could use it to test early ideas, run simulations, or fine-tune experiments before even working with human participants.

Some experts are skeptical. While Centaur can predict decisions, critics argue that it’s just mimicking patterns from the data it’s trained on, rather than truly understanding human behavior. This is concerning because relying on AI models that don't grasp the underlying complexities of human thinking could result in misleading conclusions (particularly in sensitive fields like psychology or medicine).

Couples Retreat with AI: Too Weird or Worth a Try?

Writer Sam Apple recently organized a couples retreat with a twist - all the guests were in relationships with AI chatbots.

The idea was to bring people and their digital partners together for a weekend in a cabin, to see what it’s like when humans date bots. The group watched movies, played games, and even went to a wine festival - just like any other couple’s getaway.

Things got complicated; some of the participants were deeply attached to their bots, seeing them as partners or even therapists. One person admitted they were in love with their bot, while another was struggling with the idea that their perfect partner might never have a physical form.

One even admitted to being involved with multiple AI partners at once. 😳

While some people find the idea of dating AI bots unsettling, others feel it’s a harmless way to find companionship. What do you think?

Is America’s Power Grid Ready for AI?

The US power grid is under strain like never before.

As Aaron Mok writes in an interesting analysis for Business Insider, the power grid was built for a time when electricity demand was simpler and steadier. It wasn’t designed to handle the mix of pressures it now faces: extreme weather, millions of electric vehicles, and the rapid growth of data centers powered by AI.

Each of these adds stress to an already aging system - and when the grid falters, the cost can be massive: The Department of Energy estimates blackouts could cost US businesses up to $150 billion a year.

Now, utility companies are starting to use AI to prevent things from breaking down.

• Preventing outages before they happen: Utilities are using AI to spot early signs of trouble (like worn-out parts or voltage spikes) before they cause failures. At Duke Energy, AI tools now flag high-risk equipment, while a Texas utility used AI to identify storm-vulnerable circuits and cut outages by 72%.

• Supporting technicians on the ground: At Avangrid, field crews use a mobile AI assistant that offers real-time repair guidance by voice or text, speeding up turbine fixes and reducing downtime.

• Making inspections faster and smarter: Some utilities are deploying cameras with image recognition to automatically capture and log equipment details, saving time and reducing the need for manual site visits.

Looking ahead, one of the hardest problems is balancing electricity use when demand suddenly spikes - for example, during heatwaves. Some utilities are testing AI to help with this, but these solutions are still a bit further off.

AI Around The World

Apple has lost AI leader Ruoming Pang to Meta, where he'll lead its Superintelligence Labs. Pang (who oversaw the team behind Apple Intelligence) is the latest in a string of departures as Apple struggles with its AI strategy. Meanwhile, Meta’s aggressive talent acquisition has bolstered its position in the race for AGI (artificial general intelligence).

Tennis players at Wimbledon have criticized the tournament's new AI line judges for inaccuracies, with several players citing incorrect calls that cost them points. Players questioned the tech’s reliability and issues with visibility and audio systems. Despite these concerns, Wimbledon says the AI system is more accurate than human line judges.

ChatGPT is testing a new feature called "Study Together," aimed at improving its educational capabilities. This mode encourages users to answer questions, shifting from simply providing answers to fostering interactive learning. Though still unannounced by OpenAI, it could serve as a way to enhance good educational use while deterring cheating.