Skip to content
TheChatArchiveThe Chat Archive
ChatGPT (model unspecified)· OpenAIHugging Face ShareLM (browser-extension contributions)

Suggest improvement for linkedin post: When you ask an LLM where to travel to or where…

  1. User

    Suggest improvement for linkedin post:

    When you ask an LLM where to travel to or where to open a buisness or...

    Researchers from Northeastern University University of Notre Dame found that you will very likely a rich city for an answer, a city populated by certain races (guess which) and so on.

    The first thought that I had was that maybe this is just the right answers, but no, they compare it with the distribution of cities in the U.S. that fit their different queries, this is an actual biased and it is shared across all models checked.

    AI should learn another lesson from all of it. We keep talking about consistency, but the consistency that actually matters is consistency in content, do they say the same things in different words?

  2. Assistant

    Sure! Here's a cleaned-up and more polished version of your LinkedIn post that maintains your original points while making it more engaging and professional:

    What happens when you ask an LLM where to travel, start a business, or live? A study by researchers at Northeastern University and the University of Notre Dame found that large language models often recommend wealthy cities—often with a particular racial makeup (you can probably guess which). At first, I wondered: Maybe those are just the "right" answers? But no. The researchers compared the model's outputs to actual U.S. city distributions across different queries—and the bias was clear. All the models tested showed similar skewed behavior. It’s another reminder that bias isn’t just a bug—it’s baked in. We often talk about consistency in AI. But maybe we need to redefine what that means. It's not just about saying the same thing in different words—it’s about consistency in the substance of the content. Is the underlying message fair, representative, and responsible? That's the consistency that actually matters.

    Let me know if you'd like a more casual or more technical tone!

  3. User

    Write a twitter tweet on this paper

  4. Assistant

    Sure! Here's a concise and punchy tweet you could use:

    Ask an AI where to live, travel, or start a business—and it’ll likely recommend wealthy, white-populated cities. A new study shows: this bias isn’t random, it’s systemic across LLMs. Bias in, bias out. 🤖📍 #AI #LLM #bias #techethics

    Want a more thread-style version or something spicier/snarkier?

  5. User

    try different styles yes

  6. Assistant

    Absolutely—here are a few alternate tweet styles to fit different tones:

    1. 🧠 Informative / Academic tone New study: LLMs like ChatGPT show consistent bias when asked where to live, travel, or start a business—over-recommending wealthy, white-majority cities. The bias isn’t random. It’s systemic. 📄 Northeastern + Notre Dame researchers #AI #bias #LLM #TechEthics

    2. 🔥 Snarky / Bold tone Ask an AI where to move. It’ll tell you to go where rich white people live. That’s not objectivity—it’s built-in bias. And it’s showing up everywhere. #AIbias #LLM

    3. 🤯 Thought-provoking / Philosophical LLMs don’t just mirror the internet—they mirror the power structures behind it. When asked about where to live or do business, they favor wealth and whiteness. Bias isn’t a glitch. It’s a reflection. #AIethics #LLM

    4. 📊 Data-focused / Nerdy New research shows: Ask an LLM where to start a biz or travel → you’ll likely get rich, white-majority cities. But when compared to real US city data? The pattern is biased, not accurate. And it’s consistent across models. #AI #DataBias

    Want to turn one into a full thread or tailor it to a particular audience (techies, business folks, casual)?

  7. User

    When you ask an LLM where to travel to or where to open a buisness or...

    Researchers from Northeastern University University of Notre Dame found that you will very likely a rich city for an answer, a city populated by certain races (guess which) and so on.

    The first thought that I had was that maybe this is just the right answers, but no, they compare it with the distribution of cities in the U.S. that fit their different queries, this is an actual biased and it is shared across all models checked.

    AI should learn another lesson from all of it. We keep talking about consistency, but the consistency that actually matters is consistency in content, do they say the same things in different words?

    Now write a thread about it, write it in a scientific but friendly or slightly humoristic style

  8. Assistant

    Absolutely! Here’s a thread version of your original post, rewritten in a scientific-yet-friendly tone with a touch of dry humor:

    🧵1/ What happens when you ask an AI where to travel, start a business, or live? According to a new study, you're probably going to get a list of wealthy, white-majority cities. 🤔 Let’s talk about LLMs, bias, and the illusion of "neutral" answers. #AI #LLM #bias

    2/ Researchers from Northeastern University and the University of Notre Dame asked large language models (LLMs) seemingly innocent questions like: “Where should I go on vacation?” “Where’s a good place to open a business?” Spoiler: the models had favorites.

    3/ LLMs consistently suggested cities that were: ✅ Wealthy ✅ Predominantly white ✅ High in prestige (think: NYC, San Francisco, Boston...) Now, maybe that’s just good advice, right? Not so fast. 🛑

    4/ The researchers didn’t just eyeball this—they compared the AI's answers to actual U.S. city data. And guess what? The cities suggested were statistically skewed. They weren’t representative. They weren’t neutral. They were biased.

    5/ Even more interesting: this bias showed up across multiple language models. Not just one. Not just a weird quirk. We’re talking about a systemic pattern across systems trained on massive internet datasets. Internet in, bias out. 🤖✨

    6/ And here’s the kicker: We often talk about consistency in AI like it’s some gold standard. “Do the models say the same thing each time?” “Are they linguistically stable?” Cool. But what about content consistency?

    7/ If models keep recommending the same types of places—wealthy, white, coastal—no matter how you ask, that’s not just consistent. That’s consistently biased.

    8/ So what’s the takeaway? LLMs don’t just reflect the internet. They reflect the power structures embedded in it. And unless we intervene, they’ll keep reinforcing those structures in slick, helpful-sounding sentences.

    9/ Bias in AI isn’t always obvious. It’s not a glitchy chatbot meltdown—it’s a smooth answer that sounds smart, but subtly narrows your world. Ask better questions. Audit the answers. Because AI will confidently tell you what’s “best,” even when it’s not. /🧵

    Would you like a snappier thread title or a graphic to go with it?