Reality check for anyone trusting AI on long tasks: A new PNAS Nexus study gave top models the classic Stroop attention test. Short lists: 90%+ accuracy. Long lists: accuracy collapsed toward zero. The lesson isn't 'AI is dumb' — machine attention degrades with length in a way ours doesn't. That's what bites in long agent runs. Where have you caught AI losing the thread? More at www.contentbuffer.com #AI #MachineLearning #LLM #TechNews #Research