am I on the hook for a Private Jet?

Short summary

A startup founder recounts how his interns created the 'PJ Branch' (a secret codebase) to pursue an impossible challenge: achieving #1 on Terminal-Bench 2.1, the most competitive AI benchmark. They succeeded with 84.3% accuracy—the highest ever recorded—using an older model while frontier labs competed with their latest. Now the co-founders seriously consider honoring their joking promise of a private jet reward.

• •Interns achieved 84.3% on Terminal-Bench 2.1—the highest score ever recorded
• •They outperformed frontier labs using a previous-generation model instead of the latest
• •Co-founders face a genuine moral and business decision about honoring their joking bet

Generated with AI, which can make mistakes.