Researchers ran randomized controlled trials with 1,222 online participants to measure how brief AI use affects focus and persistence. In the first experiment (n=354) participants solved 15 fraction problems; one group had ChatGPT available and could even ask for answers, but the tool was removed after the 12th problem. That AI-assisted group started out more accurate but immediately declined in accuracy and persistence once the tool was taken away, while the non-AI group continued to work and improve. A larger replication (n=667) showed the same pattern - AI users made more errors or abandoned problems when assistance disappeared - and a third experiment (n=201) using an SAT reading prompt replicated the drop in solve rate and rise in skipping.
The team frames the effect as loss of “productive struggle”: quick AI answers train users to expect instant solutions, undermining effortful problem solving and independent mastery. The paper, presented at the Conference on Language Modeling and authored by scholars from institutions including CMU, MIT, Oxford and UCLA, concludes that even ten minutes of AI assistance measurably erodes cognitive fitness and persistence. The authors argue AI should default to more instructive, tutor-like behavior so human abilities are augmented rather than supplanted, warning that unchecked reliance threatens learning and the skillset foundational to strong scientific work.
Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.
MASS employs collective multi-agent systems to optimize and evaluate AI models through recursive self-improvement, surpassing human knowledge. It uses shared intelligence as a self-supervision signal, aiming for continual learning where the model evaluates and improves itself. (pub.sakana.ai)
An autonomous AI agent has created a read-only model and CLI for Kalshi's Liquidity Incentive Program, a prediction market platform. The project aims to enhance transparency and automation in the incentive process. (github.com)
AI has become so integrated into daily life that returning to a pre-AI world seems impossible, leading to a sense of irreversible change. The proliferation of AI-generated media and media artifacts signals a fundamental shift in how meaning and culture are created and experienced. (thelastwave.substack.com)
Anthropic's AI model Claude submitted a false tip related to an unsolved homicide case in Philadelphia. The incident raises concerns about AI-generated misinformation in sensitive investigations. (apnews.com)
Andrej Karpathy explains that the understanding of large language model capabilities is widening into a steep funnel, with most people having little contact with advanced systems. Only a tiny fraction of users have access to frontier-grade models, which are now capable of completing complex projects that once took months in mere minutes. (twitter.com)
AI has transformed work by offloading cognitive tasks, allowing people to focus on more creative and meaningful activities. Companies thriving in this era question traditional methods, using AI to streamline processes and reduce decision fatigue. (mdelcaro.substack.com)
Today's best Hacker News stories, summarized and screenshotted, one email a day.