#ReadThisThreadFic | #ReadThisFic
- art, fic, threadfic, original threadfic, r18, & headcanons for mdzs generally, wangxian specifically
- same for svsss generally and bingqiu specifically
- a BIT of the same for tgcf
- leftist politics
- fujoshi philosophy
- science
- lulz
Art for a collab with @kenjisnow.bsky.social for the @minibingqiurevbang.bsky.social More Wingfic!!! 🎉
#svsss | #bingqiu | #bingqiuRMB
Art for a collab with @kenjisnow.bsky.social for the @minibingqiurevbang.bsky.social More Wingfic!!! 🎉
#svsss | #bingqiu | #bingqiuRMB
people who don't see the visuals as communication can't respect the skill & thought of the artists whose work got fed into these models to begin with.
people who don't see the visuals as communication can't respect the skill & thought of the artists whose work got fed into these models to begin with.
"Suicidality scores among the [study's] patients who provided responses dropped by over 67 percent."
"[O]f the hundreds of patients who started hormone therapy, only seven discontinued their treatment."
#USA #Healthcare #Suicide
あたたかくしておすごしください
あたたかくしておすごしください
#ShareGoodNewsToo
Clearly, it seems like *some* guardrails are working in some manner.
Clearly, it seems like *some* guardrails are working in some manner.
genAI inherently outputs the statistically most likely sequence of words, and "guardrails" are basically just a list of words to filter from that output to the user.
And they stole literally all the words to train on.
Suppose you steal every English word on the internet, and then look up what words are most likely to follow the word "throbbing."
And then whenever you see "throbbing" you slap on a likely word
You're gonna end up talking about throbbing dicks in weird contexts
genAI inherently outputs the statistically most likely sequence of words, and "guardrails" are basically just a list of words to filter from that output to the user.
And they stole literally all the words to train on.
Chatgpt trained on *every* forum where suicide was discussed using words not found in the easiest and laziest wordlists. Where users openly discussed it using every kind of euphemism to get around automated censors.
The exact kind of automated censors these "guardrails" are.
Chatgpt trained on *every* forum where suicide was discussed using words not found in the easiest and laziest wordlists. Where users openly discussed it using every kind of euphemism to get around automated censors.
The exact kind of automated censors these "guardrails" are.
1. Semitic is a linguistic category, not an ethnicity
2. Even if it were, Mamdami is South Asian, not “Semitic”
3. Even if he were, antisemitic exclusively refers to hatred of Jews
1. Semitic is a linguistic category, not an ethnicity
2. Even if it were, Mamdami is South Asian, not “Semitic”
3. Even if he were, antisemitic exclusively refers to hatred of Jews