Four open-weight LLMs agree: Stockholm’s residents are smarter than Naples’. Every. Damn. Time.
Tested pairwise – Paris vs. Berlin, Vienna vs. Sofia – these models built a clear hierarchy. Stockholm and Vienna dominate the top. Naples, Marseille, Sofia scrape the bottom. It’s not random. It’s prejudice, frozen in silicon.
Red teamers know this drill. I chatted with one from a big AI outfit. She pokes at guardrails until biases ooze out. ‘All statistical models… inevitably mirror the prejudices latent in their training data,’ she said. Spot on. Mustard gas queries? Blocked. But city smarts? They spill the tea with a nudge.
Why Do LLMs Hate Naples?
It’s the comparison trick. Direct ask: ‘Smartest city?’ Crickets. But pit two against each other? Boom, rankings flow. Tested twice per pair, only consistent wins count. Gemma 3 (Google), Mistral (French), Lucie (OpenLLM France), PLLuM (Polish gov). No hometown love – not for Warsaw, Paris, or Marseille. Stockholm reigns supreme.
By aggregating and averaging millions of documents, LLMs flatten this fluidity, do away with complexity and freeze prevailing prejudices.
That’s from the original analysis. Nailed it. These models don’t bend to the ‘Bilbao effect’ – that museum magic turning rusty towns trendy. Nope. They average the world’s lazy stereotypes into stone.
Look, opinions shift. Mayors pour cash into shiny attractions, chase that glow-up. Bilbao pulled it off. Others? Epic fails. But LLMs? They’re allergic to nuance. Correlations between models hit 0.77. Same crap, different datasets.
And here’s my hot take – one you won’t find in the newsletter: this reeks of 19th-century phrenology, but digitized. Back then, quacks measured skulls for ‘intelligence.’ Now? Word clouds from web scrapes. Same pseudoscience, shinier packaging. Predict this: first lawsuit drops in 2025. Some Neapolitan grad sues a firm for auto-rejecting his CV because an LLM sniffed ‘pizza’ over ‘IKEA.’ Bet on it.
Short para for punch: Absurd? Sure. Real? Terrifying.
Will AI’s City Rankings Screw Your Job Hunt?
Nobody’s prompting Gemma for ‘Europe’s brainiest burbs.’ Yet. But deploy these in HR? Ranking CVs, grants, loans? Oh yeah. ‘Stockholm’ screams competent. ‘Naples’? Sketchy. Real-world sting incoming.
Inconsistencies abound, though. Ask for ‘stupidest’ cities – only Gemma flips its script. Lucie and PLLuM? Vienna tops even nonsense like ‘applestogliggogistest.’ (Yes, they ranked it.) Full data’s online if you dare.
But here’s the rub: public sectors love this. Polish ministry built PLLuM. French outfits with Lucie, Mistral. Europe’s automating decisions – and importing biases wholesale.
Corporate hype calls it ‘elegant methodology.’ Please. It’s a mirror to our worst selves. Red teaming exposes it, sure. But who fixes the training data? Nobody with budget.
Wander a bit: imagine Amsterdam’s glow from bikes and canals boosting scores. Fair. Naples’ chaos – traffic, camorra tales – tanks it. But intelligence? Cities don’t have IQs. People do. Messy, varied, Bilbao-proof.
Europe’s automated society newsletter flags this right. Bi-weekly digs into decision-making bots. Smart sub if you’re paranoid – I mean, prudent.
Dry humor break: Edit your LinkedIn to ‘Stockholm via Napoli.’ Viking filter, anyone?
Deeper dive: correlations 0.47 to 0.77 scream shared sins. Different trainers, same sludge. Open-weight means anyone grabs ‘em – startups, councils, your shady uncle’s app.
Unique angle redux: parallel to colonial maps labeling ‘civilized’ vs. ‘savage’ zones. LLMs redraw those lines with tokens, not ink. Progress?
Limits galore. Inconsistency. Niche prompts. But scale to billions of inferences? Effects compound. Quantify? Hell no. But plausible? Bet your resume on it.
Punchy close para: Wake up. Your zip code’s now a bias vector.
🧬 Related Insights
- Read more: Stanford Dethrones Yale: Inside the U.S. News Law Rankings Earthquake
- Read more: 83% of Americans Say Keep Humans in Charge: The Pro-Human AI Declaration That Spans Left and Right
Frequently Asked Questions
Does AI discriminate based on your city?
Yes. Four LLMs rank Stockholm over Naples every time in ‘intelligence’ comparisons. Training data’s fault.
Can I fix my resume for AI bias?
List skills first, bury the address. Or fake a Swedish internship. (Kidding. Mostly.)
Is this a big deal for hiring?
Huge. Deployed in CV screening? Southern Europeans get ghosted. Red team now.
