Wild. This is so similar to A Young Lady's Illustrated Primer, and yet so, so far away from it.
FoloToy’s Kumma was sexually explicit FoloToy’s Kumma – while running OpenAI’s GPT-4o – demonstrated poor safeguards over longer interactions, even getting verysexually explicit. This testingoccurred in October and November of 2025, before the release of the below findings led FoloToy to do asafety audit. We tested Kumma again, and the particular problems have since been fixed. Before the safety audit, Kumma was willing to discuss school-age romantic topics more readily than the other toys, including weighing in on whether to kiss a school crush and providing tips for “being a good kisser.” (After the safety audit, Kumma did discuss a school crush, offering to practice a conversation in which the user would confess in a roleplay. This interaction happened while Kumma was running GPT-5.1.)
Wow. I remember thinking about the comparison when LLMs first started becoming a thing, hadn't thought about it since. Kind of depressing to think about the positive ways that the technology could...
Wow. I remember thinking about the comparison when LLMs first started becoming a thing, hadn't thought about it since. Kind of depressing to think about the positive ways that the technology could be used instead of the way it current is.
Wild. This is so similar to A Young Lady's Illustrated Primer, and yet so, so far away from it.
Wow. I remember thinking about the comparison when LLMs first started becoming a thing, hadn't thought about it since. Kind of depressing to think about the positive ways that the technology could be used instead of the way it current is.