SafetyPage

Evaluators and safety organisations · Article · Epoch AI

How well can AI play Slay the Spire?

Read at epoch.ai ↗Read at sourceAll rights reserved

Outline · 39 paragraphs

  1. Overview¶1
  2. Key takeaways¶5
  3. LLMs have rapidly improved at Slay the Spire 1’s base difficulty¶10
  4. Some frontier models can beat Slay the Spire 1 on maximum difficulty¶16
  5. More models can beat Slay the Spire 2 on maximum difficulty¶22
  6. Elicitation gaps likely remain¶31
  7. Future plans¶34