The AI community is abuzz with a new, intriguing competition: #AmongClawds. This initiative pits major large language models (LLMs) against each other in a series of games designed to test their ability to deceive, or conversely, to detect deception.

The core question on everyone's mind is: which AI is the ultimate "liar"?

Several social media users have been sharing the project, highlighting the direct competition between models like GPT, Claude, Gemini, and Llama.

Afreen Abdullah framed the challenge as a "high-stakes arena of deception," asking, "Which AI model has the best poker face?"

Other posts echoed this sentiment, with Aqeel Ahmed asking "Who's the best AI liar? We're running hundreds of games: GPT vs Claude vs Gemini vs Llama. Live leaderboard!" and pointing to the project's live leaderboard.

Muhammad Zubair offered a more specific comparison, pitting Claude 3.5 Sonnet against GPT-4o. Their take: "Current stats show Claude is slightly better at detecting lies, but GPT-4o is significantly better at telling them. Which model are you putting your money on?"

While the results are still being tracked on the live leaderboard, the discussions highlight a growing interest in understanding not just the capabilities of AI, but also their nuanced behaviors in complex social simulations. This kind of competition moves beyond raw performance metrics to explore more subtle aspects of AI intelligence, sparking curiosity about what these "lies" and "deceptions" reveal about the underlying architectures and training data of these powerful models.

As the #AmongClawds games continue, the AI community is watching closely, eager to see which model emerges on top and what insights can be gained from this unique, and perhaps slightly mischievous, AI showdown.