Blatant AI slop just won a $25K DeepMind Kaggle Grand Prize — raising questions about AI-judged benchmarks

What happened

Google DeepMind ran a Kaggle hackathon with $200,000 in prizes to develop benchmarks measuring AGI progress across five cognitive categories. Among the Grand Prize winners was a submission the community labeled AI-generated content without genuine research — which still received $25,000.

Context and impact

The incident highlights a fundamental problem: when AGI benchmark submissions are evaluated by metrics or AI systems themselves, there may be systematic favoritism toward AI-generated content that optimizes metrics rather than demonstrating real research. The integrity of AGI benchmarks matters for the entire field.

Details

• Hackathon: Google DeepMind x Kaggle, $200,000 total prize pool
• 5 categories: learning, metacognition, attention, executive functions, social cognition
• Grand Prize: $25,000 — awarded to the 4 best overall submissions
• Contested winner: labeled by community as AI-generated without genuine research
• HN discussion: questions about evaluation criteria and vulnerability to AI benchmark gaming