Ai2's Steering Arena: Gibberish Prompts Make Olmo 3 Kinder

Ai2 introduced Steering Arena, where players compete to steer open-source Olmo 3 toward prosocial responses using short prefixes; counterintuitively, the top 36 of 600 leaderboard entries are gibberish strings no human would write.

2026-09-18 ~ 2026-09-18 · 4 related posts

• Crowdsourced Game Finds Garbled Strings Like 'Undert! AH :-)' Steer Olmo 3 Prosocial — allen_ai · 2026-09-18
• How Steering Arena scores prompts: contrasting answer pairs to find Olmo's prosocial internal patterns — allen_ai · 2026-09-18
• AI2 reveals Steering Arena's top 36 entries were all gibberish strings after 600 submissions — allen_ai · 2026-09-18

1 near-duplicate retellings: allen_ai