Generative AI models are constantly evolving and surpassing capabilities previously thought impossible. One such model, the GPT-4, recently demonstrated its impressive skills in the popular game Minecraft.
With its complex behaviors and unguided learning abilities, the AI became a formidable player in Minecraft. It surpassed the previous state-of-the-art by unlocking key tech tree milestones at an astonishing speed. This was achieved through three fundamental tools: an automatic curriculum for exploration, a skill library for crafting and fighting, and an iterative prompting mechanism that generates code representing actions within the game.
What makes this AI stand out is its ability to learn new skills and use them to achieve its goals quickly. Rather than being trained on what is required to succeed in Minecraft, Voyager, as it is called, was able to learn on its own and explore new things. Its in-context lifelong learning capabilities enable it to quickly generalize already-learned skills to succeed in new contexts.
While some may view the development of an AI that plays Minecraft as a mere novelty, it has the potential to create a starting point for the development of broader generalist AI agents. The ability to learn, recall and use knowledge is crucial for any independent actor in any context, even beyond gaming.
While Minecraft is a relatively straightforward game, Voyager’s success is an example of the potential of AI in creating and learning new skills, and its vast applicability in different areas of development.
Looking forward, it is interesting to consider the impact of new AI models and their widespread influence in all industries, particularly as the technology continues to advance. The key takeaway is that we need to be aware of these advancements and potential implications for our jobs.
