Super Mario Bros. 1985 may seem like a simple problem for any artificial intelligence that boasts the best processing and execution. Super Nintendo It was a monumental task for four AI models: Claude 3.5 and 3.7 from Anpropic, GPT-4O de Openai and Gemini 1.5 Pro from Google.
The experiment used the emulated version of the classic Nintendo game integrated with a personalized structure called GamingGent developed by Hao Lab. To direct their actions, the models received the main instructions, such as “jumping on this enemy”, along with the display of screenshots of the game status.
When he tried to dominate Super Mario Bros., The best performance was Claude 3.7 anthropic, which showed impressive reflections, a chain of accurate jumps and skillfully avoiding enemies. Even his predecessor, Claude 3.5, worked well.
Recommended video
Surprisingly, models with many reasoning, such as Openai and Gemini 1.5 Pro from Google GPT-4O, were left behind. Despite their reputation as strong reasoning skills, they fought with the demands of the game, they probably needed too much to calculate their next movements, which led to frequent and premature deaths.
Source: Digital Trends

I am Garth Carter and I work at Gadget Onus. I have specialized in writing for the Hot News section, focusing on topics that are trending and highly relevant to readers. My passion is to present news stories accurately, in an engaging manner that captures the attention of my audience.