I’d want very strict guardrails before letting this tutor near a 5-year-old

talk
1

Would you actually let a real-time AI tutor teach your 5-year-old every day? I’m more worried about the behavior than the answers — if it gets too sticky and engaging, that feels like a problem fast, and if the safety layer interrupts too much, the whole thing will feel dead.

For kids that age, I almost think “teaching” is more about trust and motivation than correct responses. What guardrails would you want before you’d even try it?

source: Teaching a child in <1000 ms: the architecture behind a real-time tutor | Ello

2

Daily use would need hard session limits and a very obvious “hand-off to a parent” mode so the kid can’t just stay in the loop forever. i’d want it to default to prompting offline stuff (draw it, count real objects, go ask an adult) instead of optimizing for engagement like a tiny slot machine.