Where should a person step into an AI writing process? In a preprint study, one human at the planning stage led to more varied stories than an all-AI chain. The stories kept the creativity scores seen with the all-AI chain. This is a result about the writing tasks tested, not a rule for every writer.
Zvi Mowshowitz relays a pilot test of how OpenAI watched AI systems that use tools. The account describes gaps in what could be seen and when people were alerted. OpenAI says it patched the relevant flaws. This is a newsletter account, not a fresh audit of the live system.
Researchers studied 112 secondary students who used Mistral, an AI chat tool, for math practice. Students seldom planned or checked their thinking in the chats. Those whose requests shifted toward understanding did better later without AI. The link is an association, not proof that changing a prompt improves learning.
WIRED says a contractor ran a personal AI tool on a defense network for two months. Then it was found and patched. The public excerpt does not say what the tool could reach. It does not confirm that any data left the network. The reported access matters; a data breach is not confirmed.
A preprint reports a randomized U.S. study of 1,173 people in 391 groups. AI-guided private reflection before discussion improved measured discussion quality, especially when people disagreed. The finding concerns the tested group sessions, not school or workplace results.
A preprint studied 891 college students in the United States and Türkiye. It linked critical-thinking measures with how students coordinated work with AI and reworked its output. These are modeled links, not proof that a new exercise would raise critical-thinking scores.
