GPU World
Points and comments are a snapshot, not live.
A contest asks writers to imagine a world where everyone has access to frontier AI by 2040.
The GPU World contest invites participants to envision a future where AI progress stops as of September 2026, but hardware scaling makes the equivalent of a B300 GPU for contemporary LLMs available to every human by 2040. The premise assumes no superhuman AI, just faster and cheaper frontier models like Fable or Sol. Questions posed include impacts on surveillance, education, social media, healthcare, and the developing world. The contest offers a $40,000 grand prize and is judged by Neal Stephenson, Gwern, and Matt Huang. LLM use is permitted but discouraged, with disclosure requested.
What commenters are saying
Commenters debated the plausibility and implications of the premise. Several noted the energy challenge: 8 billion GPUs at 1400 watts each would demand 11.2 terawatts, far exceeding current global production of 3.6 terawatts. Others pointed to specialized chips, analog computing, and model optimization as potential mitigations. One commenter praised the premise for avoiding cynicism, while another referenced Adrian Tchaikovsky's 'Service Model' as relevant fiction. A brief exchange touched on the Butlerian Jihad from Dune as a cultural reference.