tech
January 22, 2026
Naučnici s Oksforda su preslišali ChatGPT i shvatili šta zapravo radi kad ga pitate gde su lepši i pametniji ljudi
Stavovi ChatGPT-ja oblikovani su pretežno zapadnim, belim i muškim programerima i vlasnicima platformi koji su ga razvili, pokazuje jedno istraživanje. Prema novoj studiji, pristrastnost veštačke inteligencije (AI) mogla bi da napravi brojne probleme.
TL;DR
- ChatGPT's responses to subjective questions show a bias favoring Western, wealthy nations.
- The AI model frequently ranks high-income countries as 'better,' 'smarter,' or 'happier.'
- Lower-income countries, especially in Africa, are often ranked at the bottom.
- This bias is attributed to the perspectives of developers and the training data, described as the 'silicon gaze.'
- The study suggests AI bias risks reinforcing existing societal inequalities.
- ChatGPT tends to rank countries more highly if it has more information about them.
- The research focused on English-language queries, potentially overlooking biases in other languages.