Large Language Models

“We do not sell products. We build experiences.” If that sounds like a chatbot, there’s a name for it. A new preprint examines why AI models use epanorthosis, a self-correcting figure from classical rhetoric.

Google’s AI Overviews provide direct answers to billions of searchers, but a new analysis reveals that one in nine generated claims lacks source support. The system also bypasses traditional search rankings, threatening the revenue of original web publishers.

An algorithm can’t feel the sting of losing money, but people will sacrifice their own cash to punish one anyway. In experiments with more than 6,000 participants, people retaliated against unfair machines just as often as against unfair humans.

People going through their first episode of psychosis may be especially vulnerable to chatbots that validate distorted beliefs, a new study suggests. At one large medical center, 61% of “AI psychosis” cases involved first-episode patients.

Chatting with a chatbot seems easy, but a small brain imaging study suggests six-year-olds don’t yet coordinate their brain’s control networks the way adults do during the conversation, which the researchers say is why adult guidance matters.

Compared to humans, language models like GPT-4 are highly optimistic about the creation of human-level AI. Researchers suggest this constant, gentle push of machine positivity could slowly alter how the public perceives artificial intelligence.

AI models are judging people by their looks. A new study reveals that popular AI chatbots rely on arbitrary facial features to determine if someone is trustworthy, competent, or even likely to commit crimes.

A new study of healthy adults and patients with severe aphasia shows that the human brain uses distinct networks for processing language and solving complex logic puzzles, challenging the long-held idea that we think in words.

Artificial intelligence models are highly susceptible to subtle changes in how options are presented. A recent study published in the Proceedings of the National Academy of Sciences provides evidence that this hypersensitivity leads to easily manipulated, unpredictable outcomes.

While popular artificial intelligence programs demonstrate impressive text capabilities, new research shows they still fail basic visual and numerical logic tests. The findings reveal a wide gap between language memorization and true foundational reasoning in modern algorithms.

When tested with a classic psychological assessment, advanced AI models experienced a total breakdown in focus. A new PNAS Nexus study suggests these systems lack the human-like executive control necessary to override automatic responses and maintain complex goals.

Large language models outperformed neuroscience experts in predicting study outcomes, achieving 81.4% accuracy versus 63.4%. Their success stemmed from superior contextual integration.

Most conversational large language models tend to produce left-leaning responses to political questions, with this bias influenced by fine-tuning, while base models generally lacked coherent political preferences.