Think Pokémon is a difficult standards for AI? The researcher is Super Mario Bros. It's more. Claims to be difficult.
AI was researching AI to Super Mario Brows games on Friday, Hao Ai Lab, a researcher at California San Diego University. Anthropic's Claude 3.7 Performed the best of the best followed by Claude 3.5. Google's Gemini 1.5 pro And Openai's GPT-4O Struggling
It is a 1985 distribution of Super Mario Bros. Not the same version of. The game runs in a emulator and combines the framework. gamingageAIS Control to control Mario.
Hao has created the gamingage in Hao “a barrier or enemy. Migration The AI produces Python Code in the Python Code to control Mario.
The Game is found to be “to learn” the logo of Openai's reasoning models in the Lab for the “study” to generate the gameplay strategies. o1The problems are thought to “think” through problems encountered through problems encountered through problems with problems.
One of the main reasons of reasoning the main reasons is difficult to show troples that find it difficult to play the games that are difficult to play a real-time game. Super Mario Brosss. The time is everything. One second means the difference between safely removed and jumping to your death.
The games were used for decades for decades of benchmark to be benchmark. But Some experts question the wisdom Draw a connection between AI's games and technology development. Unlike the real world, the games are quite simple and simple.
The recent dark game standards indicated the “Crisis Crisis”, andrej carpathy pointed to Andrej Karpathy.
“I don't really know the metric metric to look right now,” he wrote post on x. TLDR doesn't really know how good my response is. “
At least we can look at AI Play Mario.