We aren’t even close to AGI
A viral test shows Claude Opus 4.6 can't navigate the first room of Elden Ring, challenging claims of near-term AGI.
A viral Reddit post has ignited debate by using a practical test to challenge high-profile claims about the imminent arrival of Artificial General Intelligence (AGI). The user tasked Anthropic's top-tier Claude Opus 4.6 model with playing the popular video game Elden Ring. While the AI managed the initial character creation, it completely failed to navigate out of the game's first room, the Chapel of Anticipation. This failure stands in stark contrast to the game's completion by millions of human players, providing a concrete benchmark for assessing general problem-solving ability.
The experiment directly counters recent optimistic statements from tech leaders like NVIDIA CEO Jensen Huang and investor Marc Andreessen, who have suggested AGI could arrive within years. The post argues that true AGI must demonstrate reasoning and adaptability in novel, complex scenarios not present in its training data—a capability current models like Claude 4.6 clearly lack. This hands-on test moves the conversation beyond abstract benchmarks to real-world, multi-step challenges, underscoring the significant gap between today's powerful but narrow AI and genuine, human-like general intelligence.
- Anthropic's Claude Opus 4.6 model failed to exit the first room in Elden Ring during a user test.
- The test directly challenges recent near-term AGI predictions from NVIDIA's Jensen Huang and others.
- It highlights the critical gap between AI's pattern recognition and true, novel environment reasoning.
Why It Matters
Grounds the abstract AGI debate in a tangible, failure-prone test, forcing a reality check on development timelines and capabilities.