WIRED reports that an ad company can use a chatbot conversation up to the ad request to choose an ad. Its claims about clicks and privacy come from the company. The report does not independently prove those claims, and its partner apps do not yet have AI shopping assistants that act on their own.
A new paper studies who holds power in AI safety tests. It looks at 12 evaluation projects in Africa and the roles of ten safety institutes in three joint exercises. Helping design a test is not the same as controlling the system being tested or deciding what gets published.
Zvi Mowshowitz puts workplace accounts about slow AI adoption beside an Anthropic filing that raises the possibility of much faster AI research. The accounts do not measure adoption across organizations. The filing does not say a self-reinforcing jump in research speed has happened.
Researchers studied 260 team projects across ten semesters. Later code grew to about three times the size and became more complex, with more handling for errors. Comments did not keep pace relative to code size. The authors did not directly observe AI use, so they cannot say it caused the changes.
WIRED reports on an announced smart ring pitched as a remote control for an AI assistant that can carry out tasks. The product claim is the story here. Smooth approvals and successful results have not been independently established in WIRED’s report.
In a classroom comparison with 65 secondary students, idea-level discussion was more common with the Maike chatbot. Students who moved from those ideas to essay changes also tended to gain points. The abstract does not show that the chatbot caused higher scores between groups.
In a classroom comparison with 65 secondary students, idea-level discussion was more common with the Maike chatbot. Students who moved from those ideas to essay changes also tended to gain points. The abstract does not show that the chatbot caused higher scores between groups.