Same Model, Different App: VS Code Fumbles, Codex Nails It
dfinke · x · 2026-08-25
A test using the same LLM and complex prompt revealed significant performance differences between apps: VS Code fumbled the instructions, while the Codex app executed them perfectly, highlighting the impact of the 'harness' on model behavior.
More from Apps
- MUDD World: An AI council to test if coherent culture can emerge — __hymn · 2026-08-25
- Gradio Launches Workflow: Visual Canvas for AI Pipelines with One-Command Deploy — Gradio · 2026-08-25
- Jan's self-hosted models integrated into workflow for six months — pjausovec · 2026-08-25
- Tool to support BYOA and host custom agents — tekbog · 2026-08-25
- Using ChatGPT to analyze salary history and uncover hidden income opportunities — Med1_Ai · 2026-08-25
- New Agent Platform Offers Full Computer Control and Custom Agents — basedjensen · 2026-08-25