Home / Technology / AI Nears Perfection on Humanity's Ultimate Knowledge Test
AI Nears Perfection on Humanity's Ultimate Knowledge Test
30 Mar
Summary
- AI systems are months away from scoring 100% on Humanity's Last Exam.
- The challenging HLE tests PhD-level understanding across 100 diverse topics.
- This AI milestone signifies a rapid advance in machine intelligence capabilities.
Artificial intelligence is poised to achieve full marks on Humanity's Last Exam (HLE) within months, a testament to rapid advancements in AI capabilities. HLE, designed by tech leaders, features 2,500 questions demanding PhD-level understanding across a hundred subjects. Initially scoring as low as 3%, AI models like Google Gemini have shown significant improvement, reaching 45.9% recently.
The test was created by Scale and the Center for AI Safety to evaluate AI's breadth of knowledge and depth of reasoning. It was compiled from thousands of questions submitted globally, with a focus on those not easily found online. Achieving 100% on HLE would mark AI as a 'universal expert' and necessitate future tests on questions beyond current human knowledge.
While AI demonstrates impressive progress, experts note that specialized human fields like surgery and nuanced skills such as judgment and creativity will likely remain areas of human expertise. The development mirrors past AI milestones, such as Deep Blue's chess victory, and suggests a new era in AI's capacity to master complex intellectual challenges.
