Back to feed

Changelog Update

  • Improved prompting for both GPT-4 and GPT-4 Turbo.
    • Far fewer edit errors from GPT-4 Turbo (gpt-4-1106-preview).
    • Significantly better benchmark results from the June GPT-4 (gpt-4-0613). Performance leaps from 47%/64% up to 51%/71%.
  • Fixed bug where in-chat files were marked as both read-only and ready-write, sometimes confusing GPT.
  • Fixed bug to properly handle repos with submodules.