I agree embedding the titles into a LLM’s latent space is a sound technique, and that it lets you measure shifts in title content in an objective way, but your conclusion that it’s becoming “extreme” seems like editorializing. In particular, you lumping in fauxcest with rape and torture as “sexual violence” strikes me as, … well, I’d say bizarre, but at the very least, subjective.
You could just as well conclude that a shift from “lonely housewife fucks handyman/delivery boy” to “help me, stepbro, I’m stuck in the washing machine” signifies a trend towards lighthearted whimsy.
Hi Shankar, I will concede it is editorializing: these are my conclusions based on the data. As to whether or not it is bizarre, I will repost my response to Tao Lin:
“”Incest is not a subcategory of sexual violence” is something of a loaded statement. Many “stepsister” videos highlight a certain kind of appearance and context: young, with a backpack, possibly braces, possibly in a setting in which they are still under the authority of a supervising adult (“mom and dad” etc). The implication, left unsaid, is that they are under the age of consent, which qualifies as statutory rape in America. IMO it’s sufficient justification to include it in the same category.”
You’re right that there are other, concurrent trends in tone and quality to measure. “Lighthearted whimsy”, which I might call more along the lines of “surreal”, is not necessarily in contradiction with violence.
I read the longer article you linked at the end. Never mind, I hadn’t realized this work is meant to make your case that “it’s time to put some brakes on the porn business.” Reposting my own comment about policy papers in a different context (bioterrorism from open AI[1]):
a “policy paper” is essentially a longer, LaTeXed version of a protest sign, intended to be something sympathetic congressmen can wave around while bloviating about “trusting the Science!” It’s not meant to be true.
I agree embedding the titles into a LLM’s latent space is a sound technique, and that it lets you measure shifts in title content in an objective way, but your conclusion that it’s becoming “extreme” seems like editorializing. In particular, you lumping in fauxcest with rape and torture as “sexual violence” strikes me as, … well, I’d say bizarre, but at the very least, subjective.
You could just as well conclude that a shift from “lonely housewife fucks handyman/delivery boy” to “help me, stepbro, I’m stuck in the washing machine” signifies a trend towards lighthearted whimsy.
Hi Shankar, I will concede it is editorializing: these are my conclusions based on the data. As to whether or not it is bizarre, I will repost my response to Tao Lin:
“”Incest is not a subcategory of sexual violence” is something of a loaded statement. Many “stepsister” videos highlight a certain kind of appearance and context: young, with a backpack, possibly braces, possibly in a setting in which they are still under the authority of a supervising adult (“mom and dad” etc). The implication, left unsaid, is that they are under the age of consent, which qualifies as statutory rape in America. IMO it’s sufficient justification to include it in the same category.”
You’re right that there are other, concurrent trends in tone and quality to measure. “Lighthearted whimsy”, which I might call more along the lines of “surreal”, is not necessarily in contradiction with violence.
I read the longer article you linked at the end. Never mind, I hadn’t realized this work is meant to make your case that “it’s time to put some brakes on the porn business.” Reposting my own comment about policy papers in a different context (bioterrorism from open AI[1]):
Not to be confused with the duplicitously-named OpenAI.