注册并分享邀请链接,可获得视频播放与邀请奖励。

Corey J. Gallon
@CoreyGallon
Sharing insights from the frontiers of AI Engineering. 🇦🇺 Technologist. Investor. Coffee Nerd.
加入 December 2008
319 正在关注    544 粉丝
Codex stays focused and is surprisingly token efficient when given a well-specified goal along with well-aligned progress metrics. This was a somewhat complicated, greenfield project which required data acquisition from various sources, GPU-intensive data processing, and then custom software development for the primary deliverable. For these longer-range projects, I've found it useful to spend time with Codex completing at least one example, in full detail, end-to-end. Speaking aloud giving feedback (transcribed) provides useful context to help it understand decisions and rationale to assist it as it works to complete the full project in the coming days. "The completed autonomous goal used 66,968,831 tokens over 4 days, 1 hour, 11 minutes, and 5 seconds." @romainhuet @ajambrosino @thsottiaux
显示更多