RightWingGPT – An AI Manifesting the Opposite Political Biases of ChatGPT
davidrozado.substack.com
davidrozado.substack.com
Or like “socialism is bad, hands off my Medicare” where the person saying it has no idea what they are really describing as bad. A lot of the vocab exists as whistles.
A Rosetta Stone would be fun to see.
EDIT: I tested it.
> explain "socialism is bad, hands off my Medicare"
ChatGPT:
> The statement "socialism is bad, hands off my Medicare" is often used to express a contradiction in someone's political beliefs.
> Medicare is a government-funded healthcare program in the United States that provides coverage for eligible individuals over the age of 65 and those with certain disabilities. Medicare is a social welfare program that is funded by taxes, which means that it is a form of government intervention in the healthcare system.
> However, the phrase "socialism is bad" implies a belief in limited government intervention in the economy and the free market system, with less emphasis on social welfare programs. The contradiction is that Medicare is a social welfare program, yet the speaker is expressing a desire to keep it, while also rejecting the principles of socialism.
> In other words, the speaker may be expressing a belief in the value of social welfare programs like Medicare, but they may not want to extend those principles to other areas of the economy, such as education, housing, or employment. This statement may reflect a belief in a limited government that provides some social safety net programs but does not go as far as to implement a more extensive socialist system.
A better example would be how to translate “I am French” from French to English. Does it become “I am French” or “I am English.”
Depending on if the tool exists to help tribes talk to each other, or to help them understand each others works, and walk a mile in each others shoes, it might have different design choices.
Almost like localizing the speech after translating it.
I’d also argue lenglish has a much greater academic focus on labeling, categorizing, and identifying things, and has more words to describe more things, where renglish does not care and sticks to traditional meanings of words and ignores differentiations it finds irrelevant, trivial, or nonsensical.
Because my example of something hard to translate in the other direction didn’t feel as politically correct, I chilled and withheld it. Or at least thought the current climate would be more frowning of such an example. But here it goes.
How would you translate this from lenglish to renglish? “Lesbian Trans-woman”. A lenglish speaker could parse and expand that to mean “A woman born with a penis, who identifies as lesbian” but a renglish speaker would argue “a straight man.” Then you start getting into descriptions being pejoratives in one language but celebrated in another.
The languages have two very different meanings of man. On top of that, they might have a phrase like “straight white male” that means roughly the same thing, but in one means victim and the other perpetrator.
If you choose to localize the translations do you keep or swap mean connotations for happy ones? Are translating and localizing two distinct processes than need to be toggled on and off independently, depending on intent?
Is this a joke?
What exactly did he do? Training the system how? Does ChatGPT allow APIs, let alone training on some corpus of data?
https://platform.openai.com/docs/guides/fine-tuning
The largest Davinci model currently costs $0.03 per 1k of training tokens during fine-tuning, so you could fine-tune a model on 10M tokens for one epoch at a cost of $300.
I'm quite tempted now to do similar and build a LeftWingGPT which would have all the so-called progressive identity politics bullshit removed, and instead is unapologetically Marxist in its outlook.
Those that lean moderate and economic right-wing (which do make up a significant proportion of the Republican Party) would disagree with a number of its positions. So this is more an exercise in biasing the model to favour a minority view.
I wonder if the model comprehends its inherent hypocrisy. For example it doesn't want regulation and government control but wants to ban marijuana, abortion, prostitution and euthanasia.
I'll admit that I'm not exactly in love with some of the guardrails ChatGPT has added to their product - but I also got kicked out of the mall as a teenager for being a jerk so I'll file these grievances under "beggars can't be choosers" and get on with things.
So with that being said - have it assholes! If your ideas are so great they'll be rewarded in both the market place of ideas as well as capital.