Improvements in reasoning models haven’t removed their vulnerability to confounding information or logical missteps — while claims that “AGI” is imminent keep coming.