[chki]

I think it's reasonable to assume that ChatGPT has politically similar viewpoints to those that made the fine-tuning and those are (in general) well-educated people living in California who hold (in general) more left leaning and libertarian views. I could also imagine that the internet text corpus as a whole is somewhat left-leaning (for example, older people are more conservative and use the internet somewhat less frequently).

However, regarding this specific chart, it's important to note that the translation from asking questions into answering "strongly agree", "agree" etc. can be heavily biased. Also, tuning these compasses can be difficult. Just some things to keep in mind, political viewpoints are not hard science and colorful charts don't force them into being quantifiable.

[pavlov]

> “…older people are more conservative and use the internet somewhat less frequently”

Older people may spend all their time writing on the Internet but it’s often inside a walled garden site. Grandpa’s Facebook comments are not visible to scrapers.

[ausudhz]

Cambridge analytica would think otherwise

[kortilla]

> more left leaning and libertarian views

Left leaning sure, but absolutely not libertarian in the Bay Area. It’s a stronghold of big government ideas like UBI, state healthcare, etc. Democrats consistently receive 90%+ of the vote and the ones that do win are standard big federal government supporters.

Nobody who works on OpenAI is likely to be an out-of-the-closet libertarian. So it’s not really a surprise that it would be configured to spew orthodoxy of the California liberal.