Hacker News Reader

"OpenAI hired Kenyan company to make ChatGPT less toxic"

"Kenyan workers helped make OpenAI's ChatGPT less toxic"

"OpenAI paid Kenyan workers minimum wage to work on ChatGPT"

Many different things are objectively true, but of course they mention the pay, because they know that to most of their "Western" readers it sounds outrageous.

View on HN

The pay is not even close to the most outrageous way to headline this article: it uses "toxic" as a euphemism for, e.g., not generating CSAM, rape, and other sexual violence.

A computer cannot generate CSAM. Please do not misuse that phrase and diminish its impact.

If an AI uses source images to generate novel reproductions, wouldn't that still count as CSAM?

Regardless, I should have included the phrase "depictions of" or similar.

If an AI is generating novel images, I wouldn't call them CSAM even if they fall under a rule that makes them illegal.

If they're not properly novel then it depends on what the source images were, I guess? But it's not like you'd need to train it on anything obscene for it to figure out how to combine bad things with nudity.

Reply on news.ycombinator.com