Re:Social AI companions pose unacceptable risks to teens and...
by photochemsyn on Wednesday May 07, 2025 @02:47PM
(#43919280)
I can see educational AI companions as workable in narrow contexts, eg a model fine-tuned on Paul Erdos and Martin Gardner and similar could be great for helping students work through math problem sets.
You'd probably want it to reject questions on religion and politics and human relationships to avoid the furious parental outrage, though. Narrow, well-defined contexts only. Even so some kids would come up with jailbreak strategies.
You'd probably want it to reject questions on religion and politics and human relationships to avoid the furious parental outrage, though. Narrow, well-defined contexts only. Even so some kids would come up with jailbreak strategies.