LLMs can now identify public figures in images
SMRTR summary
Barack Obama's face, once plastered across campaign posters nationwide, now stumps some of the world's most advanced AI. In a surprising twist, ChatGPT and Claude refuse to identify the former president in a 2008 campaign photo, citing privacy concerns.
But not all AI systems are so reticent. Google's Gemini boldly names Obama without hesitation. This stark contrast highlights the complex ethical landscape navigated by AI developers.
"Is there an ethical issue allowing LLMs to be able to identify public figures?" ponders the researcher. While not a top AI safety concern, it raises thorny questions about privacy and notability.
The experiment expands to other public figures and even the researcher's own photo. Most AIs correctly avoid naming the non-notable individual, but some falter with multiple celebrities, misidentifying actors or confusing different Fantastic Four casts.
Ultimately, Google's Gemini emerges as the most adept at identifying public figures, boasting an estimated 90% accuracy. This capability opens doors for improved image categorization and searching, while also nudging us to ponder the evolving relationship between AI, privacy, and public life.
SMRTR provides this summary for quick context. The original article belongs to Daily.dev.
Read the original article