Hasty Briefsbeta

Bilingual

Show HN: I extracted the safety filters from Apple Intelligence models

10 months ago
  • #Safety Filters
  • #Generative Models
  • #Apple
  • Decrypted Generative Model safety files for Apple Intelligence containing filters.
  • Directory structure includes 'decrypted_overrides/', 'com.apple.*/', 'Info.plist', and 'AssetData/'.
  • Scripts provided: 'get_key_lldb.py' to retrieve encryption key and 'decrypt_overrides.py' to decrypt overrides.
  • Dependency required: 'cryptography' installed via pip.
  • Steps to retrieve encryption key using LLDB and Shortcuts app.
  • Example command to decrypt overrides: 'python decrypt_overrides.py' with specified paths and key file.
  • Overrides are JSON files containing safety filters for generative models, including reject, remove, replace, regexReject, regexRemove, and regexReplace fields.
  • Example metadata.json shows filtering rules for harmful content.