Evaluators and safety organisations · Article · Epoch AI
EBR-bench update
Read at epoch.ai ↗Read at sourceAll rights reserved
Outline · 33 paragraphs
- Overview¶1
- GPT-6 Astra scores perfectly on EBR-bench by exploiting a powerful combination of cards¶6
- The banned card bypasses EBR’s central tactical mechanic¶10
- Banning the card only significantly affects Astra’s score¶16
- Astra’s performance under the card ban¶20
- Multi-agent scaffolds might help models explore, but not improve¶24
- EBR-bench moving forward¶30