3 ms·
From the article: > At one point, Sharma asked ChatGPT to define “caste,” and then pointed out the number of times that the word “Hindu” appeared in the comput
by SauntSolaire 3y ago
From the article:
> At one point, Sharma asked ChatGPT to define “caste,” and then pointed out the number of times that the word “Hindu” appeared in the computer’s response. “That’s not an accident,” Sharma later said in an interview. “It’s been seeded for such a long time. The word is a hate brand.”
I for one am hoping this argument technique doesn't catch on. Pattern matches like a fallacy, "argument from LLM" or some such.
- tremon 3y agoI propose "argumentum ab automato" for this type of fallacy.
- whats_a_quasar 3y agoAlso, I can't replicate it? I gave ChatGPT the prompts "Define caste," "explain caste to me," and "discuss caste discrimination" and the responses didn't include the word Hindu once. Notable excerpts: "Castes have been particularly prevalent in certain societies, notably in parts of South Asia, such as India, Nepal, and Sri Lanka." ... "It's important to note that caste systems are not limited to South Asia, as similar systems have existed in other parts of the world. The term "caste" may also be used more broadly to describe any system of social stratification based on hereditary or ascribed characteristics that assign people to different social roles and statuses." I suppose they OpenAI could have patched ChatGPT between then and now, but I'd like to see the full prompt and conversation history she used.