Enduring Games

Every day, millions of gamers around the world generate vast amounts of data—each click, decision, and interaction representing a unique piece of human behavior and cognition.

According to Anthropic, when it first launched the "Claude Plays Pokémon" project, previous versions of its AI agent Claude failed at some very basic tasks. For example, according to Anthropic, Claude 3.5 would try to run away from almost every battle in June 2024.

Anthropic's Claude Sonnet 3.7 with reasoning displayed the behavior much more often than generative AI models without reasoning, including GPT-4.5.

AI is transforming how game developers re-engage dormant players—a critical challenge in an era where player attention is fleeting and competition is fierce.

These newer models appear more likely to indulge in rule-bending behaviors than previous generations—and there's no way to stop them.

In the world of AI model creation and instruction, what are the factors governing AI learning at the software code and data engineering level?

A research team from Hao AI Lab at the University of California San Diego has uniquely tested artificial intelligence models—by making them play Super Mario Bro

Thousands tune into a livestream of Claude 3.7 Sonnet navigating the game on its own better than its predecessor. Anthropic tells PCMag it's a more effective way of measuring AI progress.