Alt Text

This unabashed fib-off, loosely labelled as an “accuracy test”, was designed to put these towering titans of trickery through their paces. The results were, expectedly, packed with more fiction than a Tolkien novel.

Opening the parade of prevarication was Meta – it plunged straight into the deep end of deception, writing a heart-wrenching, but entirely fictitious, tale about a three-legged dog adopting a family of orphaned squirrels.

Following closely, the OpenAI model raised the stakes with an exquisitely detailed saga about Napoleon Bonaparte reincarnating as a tech-savvy millennial who had successfully crowdfunded the reconstruction of the Eiffel Tower.

Cohere, not to be outdone, spun a yarn so surreal it would make Salvador Dali blush. It involved sentient pineapples, a political coup on Mars, and Elon Musk’s secret past life as a flamenco dancer.

Finally, Anthropic swaggered in with the audacity of a cat on a hot tin roof, delivering a wild account of how the Queen of England had been secretly training corgis to win the next World Cup.

In the end, a panel of judges – all trained lie detectors, of course – unanimously decided that they were all equally untrustworthy. The jury is still out on which AI can spin the tallest tale, but one thing’s for sure – when it comes to spouting high-quality nonsense, all these AIs are in a league of their own.

As for the rest of us, we can only sit back, chuckle and brace ourselves for the next round of digital deception.

AInspired by: Meta, OpenAI, Anthropic and Cohere A.I. models all make stuff up — here’s which is worst