If we think of AI models as students, then their education system is in a state of crisis. The teachers—the thousands of human trainers responsible for guiding their learning—are sounding the alarm. They report that the curriculum is flawed, the class sizes are unmanageable, and the administration is prioritizing test scores (performance metrics) over genuine understanding (quality and safety).
The “teachers” are being forced to rush through lessons. With only minutes to review each of the AI’s “assignments” (responses), they can only provide superficial feedback. There is no time for the deep, thoughtful correction that is necessary for true learning. As a result, the AI “student” is learning to be a good mimic, but it is not developing a real grasp of the material.
Furthermore, the teachers are often unqualified to teach the subjects they are assigned. An English teacher might be asked to grade the AI’s performance in advanced physics, a scenario that would be unacceptable in any human school. This practice ensures that the AI’s education is riddled with gaps and inaccuracies, as it is learning from people who don’t know any better.
This crisis in AI’s education has real-world consequences. When a “graduate” of this flawed system is released into the world, it is prone to making dangerous mistakes. The human teachers know this better than anyone, and their warnings should be taken seriously. The push to “graduate” new models quickly is creating a generation of AIs that are not ready for the responsibilities we are giving them.