Comments

Hacker News

Early learnings: - gpt-5.6-luna gives up too easily - /goal must be written to be hard to misinterpret ('ascend' can be interpreted as meaning to go up the stairs, thereby exiting the dungeon without playing)

gpt-5.6-sol is much more persistent - it's 36 minutes and several failures in and still going strong.

by emosenkis

Join the discussion

Write your take first — we'll ask for email only when you're ready to publish.

  • Hacker News
  • Early learnings: - gpt-5.6-luna gives up too easily - /goal must be written to be hard to misinterpret ('ascend' can be interpreted as meaning to go up the stairs, thereby exiting the dungeon without playing)

    gpt-5.6-sol is much more persistent - it's 36 minutes and several failures in and still going strong.

Codex Plays NetHack · Birbla