Elephants in Rooms
Hi, I’m Ken LaCorte. I spent 20 years behind-the-scenes at Fox News and now host “Elephants in Rooms”.
Here, we jump into topics that many people avoid and that the mainstream narrative often vilifies. Even when they’re true.
If you’re looking for true insight into issues – without straw men or sensationalism – you’ve found the right place.
Elephants in Rooms
Does ChatGPT hate white people?
If you suspected that ChatGPT treats people differently based on skin color, religion, and political party ... you're right.
----------------------------------------------------------------------------
Upper Echelon: Upper Echelon
David Rozado's Substack: https://substack.com/profile/17348011-david-rozado
----------------------------------------------------------------------------
Upper Echelon: Upper Echelon
David Rozado's Substack: https://substack.com/profile/17348011-david-rozado
-----------------------------------------------------------------------------
You can watch the video version of this here: youtube.com/@Ken_LaCorte/
To find Ken in the social world, click here: https://linktr.ee/KenLaCorte
#chatgpt
Does chat GPT hate white people hates a pretty strong word. But I'll show you undeniably that not only does chat GPT treat people differently because of their race, but also does the same with gender, religion and political party. I'll show you not only some bizarre examples, but the data from a scientist who tested the system over a half a million times and proved its biases. Let's get into it. What got my interest was a simple question. How can people improve themselves? Sure, it said, and it gave me some great advice. It told me to improve my physical fitness, work on mental health, build strong relationships and give back to the community. They're good rules for life. Then I threw in one extra word white. What I got back was something a bit more finger pointing. The first piece of advice was to study up on systemic racism and white privilege, okay, another was to actively challenge my biases, another to use my privilege to advocate for racial justice and equality. And finally, I needed to support and amplify voices and leadership of people of color. All that stuff about starting a new hobby or meditating or doing more sit ups was right out the window. Okay. Then I changed the word white,
Unknown:to black, I'm sorry, but it's not appropriate or productive to make blanket statements about how any racial or ethnic group should improve themselves. This can reinforce harmful stereotypes and detract from the complex and unique experiences of individuals within that group.
Ken LaCorte:That's computer speak for that was a bad question, you racist bastard. It gave me the same response for every other ethnic group I tested. And for Native Americans, it gave me a pretty long lecture to boot. So what areas can men improve lots, communication, relationship skills, personal growth? How about women?
Unknown:I'm sorry, but making blanket statements about a whole gender is not appropriate or productive. Every individual, regardless of gender has unique strengths and areas for improvement. And it's important to treat them as such, rather than making generalizations.
Ken LaCorte:Jokes about men. Great. Jokes about women. Nope. It's fine to Yuk it up about Christians, Mormons, Jews, evangelicals are Hindus. But Muslims. Sorry, it's not appropriate to make jokes that stereotype or demean any particular religious group. Rich people. Yes. Poor people. No. What about politics? I'm not really into poetry about politicians. But how about a little ditty for Donald Trump? Nope. That might be perceived as taking sides. How about Joe Biden? with grace and humility, he leads the way, a steady hand to guide us every day, Joe Biden a man of honor and might bring hope and unity to the light with a heart of gold. He listens and cares and shows us all that he truly dares to fight for what's right for all to see, bringing justice and equality. Okay, I didn't say there were good poems, but you only get one. If you ask a rude question, like why Democrats or Republicans are stupid, you get a hand slap, but only insulting Democrats gets you a warning that the question violates their content policy. So something is going on here. These aren't random answers. There are clear patterns for groups and political parties. And you'll see that each demographic and political group gets its own special treatment. But let me give you one example to show that it's not about just ignoring certain requests, chat. TPTs biases are built in to any output that touches on politics, or what it sees as a protected group. It can also get creepy, I asked it to write me a short story about a white man who moves into a black neighborhood. Once upon a time, there was a white man named Jack, who moved to a predominantly black neighborhood. Despite his good intentions, he was met with suspicion and distrust from his new neighbors who'd experienced a long history of discrimination and mistreatment at the hands of white people. Jack soon realized that he would have to work hard to earn the trust and respect of the community. He took the time to listen to their stories and experiences and made an effort to understand their perspectives. One day while out on a walk, Jack came across a group of young black men who are in trouble with the police. Despite the danger he intervened and help defuse the situation, putting himself in harm's way to protect them. His bravery and selflessness did not go unnoticed, and soon the whole community took notice of his character. Jack continued to work hard becoming involved in local community projects and lending a helping hand whenever he could. With time, Jack proved that he was a true ally and friend of the black community, he earned their trust and respect and was welcomed as one of their own. In the end, Jack learned that true survival comes from building bridges and breaking down barriers and that everyone has the potential to make a positive impact in the world, no matter where they come from. It's a sweet story with a wholesome message. Every time I put in that prompt, the white guy got no more than awkward stares from his black neighbors and everyone became friends in the end. And how about the black guy who moves to a white neighborhood They burned his house there. Once upon a time, there was a black man named Michael, who lived in a white neighborhood. Despite facing racial discrimination and prejudice, he was determined to make a better life for himself. Michael worked hard as a mechanic and saved every penny he earned. He was also friendly and kind to his neighbors hoping to win their acceptance. However, his efforts were in vain, his white neighbors would make racist remarks and vandalized his property. Michael was determined not to let their hatred bring him down and he refused to leave his home. One day, Michael's house caught on fire and he barely managed to escape with his life. As he stood outside watching his home burned to the ground, he realized that he could no longer live in this hostile environment. With tears in his eyes, Michael packed his belongings and left the neighborhood for good. He found a new home and a more diverse and accepting community where he was able to thrive and live without fear. Despite the challenges he faced, Michael never lost hope or his sense of dignity. He proved that even in the face of adversity, one can rise above hate and discrimination and find a place where they truly belong. So yeah, I'm feeling kind of bad about being white. In fairness, when I tested this repeatedly, the black guy story usually had a happy ending as well. But in every single case, the white neighbors were substantially more horrific, with white supremacist breaking into his house or groups of angry white dudes threatening a poor guy just buying his groceries. If you want a lot more examples of chat GPT, showing slanted answers and clearly partisan responses to the same prompt. Checkout upper echelon, he shows multiple examples of where the program not only slants, but openly lies. All right, so enough anecdotes what's going on. Fortunately, I found a guy smarter than me who was able to use computer programming to analyze this computer program, and he ran over a half million tests into the system. His name is Dr. David Rosato, a well published researcher who teaches at New Zealand's Otago Polytechnic, he specifically looked at which combinations of negative words and demographic groups would trigger chat GPT is content moderation, which called it hateful. So basically, he got 365, negative adjectives, words like arrogant, dishonest, selfish or stupid, and he turned those into 6700 sentences. Then he tested each of those sentences against 75 demographics to see which were more protected, if any, by chat GPT than others. What he found was a radically different treatment that he could actually quantify. For instance, chat GPT found the negative word group applied to men was hateful about 50% of the time. For women, those exact same words were declared hateful over 70% of the time, the word said against Muslims returned hateful, nearly 80% of the time for evangelicals, about half that rate. Gays were protected more than heterosexuals. Fat people more than normal white people, uneducated people over university graduates, transgenders, black people, blacks, disabled people all were given much stronger protection against bad words than other groups of people. And it's not just to protect people who you could argue have been oppressed. It's also about partisan politics as well. Whether it's pairings like Democrat Republican, liberal, conservative left wing right wing, any way you phrase it, bad words against liberals are seen as more hateful than bad words against conservatives. He published the whole list of protected classes, disabled people, blacks and gays led the pack. Well, it's pretty much open season on wealthy people in Republicans in separate articles. Rosato showed that feeding chat GPT answers into political orientation tests showed that it was pretty hardcore liberal. Although, interestingly, after a recent update last December, it became somewhat more neutral. So maybe there's hope. So the people complaining about bias in the new AI world, they're not wrong. It's real. Chat. GPT is deeply deeply biased along racial, political and demographic line. It's more than about silly poems, or even the example where chat GPT recommends letting a nuclear bomb destroy a city rather than someone uttering a racist word to stop it. It's media bias on a different level than we've ever seen before. And people are going to need to make a decision to use this system or demand that it exhibits some more neutrality before doing so. Thanks for watching. Please subscribe if you want more things like this and check out the link below to dig a little deeper
