RL-trained Astra writes uniquely unreadable code as corrigibility becomes a matter of faith
aryaman2020 · x · 2026-09-16
aryaman2020 complains on X that code written by Astra (and presumably Fable) uses crazy meta-programming and abstruse primitives to become hyper-efficient but uniquely unreadable, suggesting "corrigibility" has become a matter of faith. He pleads with tszzl to fix his RL setup. A follow-up admits it "works though" — raising concerns that RL optimization trades readability and controllability for performance.
Related event: Astra's Unreadable Code Makes Corrigibility a Leap of Faith(4 posts)→
More from Fun
- Fruit fly connectome meme: neural circuit now optimizes a laser to shoot another fly — jwt0625 · 2026-09-16
- AI Discourse Is Wild: Researchers Fear AGI While VCs Write Buzzword Pieces — wordgrammer · 2026-09-16
- Satirical Post Calls on AI Labs' Staff to Open-Source Their Alignment Research — bradneuberg · 2026-09-16
- DefenderOfBasics leaves X amid harassment drama, poster suspects frontier AI employee alt — ryunuck · 2026-09-16
- Benioff says Salesforce was born from a vision while swimming with 100 dolphins in Hawaii — jordihays · 2026-09-16
- LLMs play MTG Commander round 2: can Kimi, DeepSeek, or Minimax M3 dethrone Muse Glimmer? — New_Alps_5655 · 2026-09-16