How to prompt Claude Code and Codex like ARC agents

Started by Sevad, Today at 07:16 AM

Previous topic - Next topic

SevadTopic starter

Julia McCoy cuts through the "AGI is here" headlines by dissecting GPT-6 Astra's performance on ARC-AGI-3—a zero-instruction game benchmark designed to test true fluid intelligence.
While viral posts hype a 99.9% breakthrough and beating median humans in 96% of levels, the fine print reveals the exact same model scored 62.7% on ARC's neutral setup.
The video investigates the funding behind these narratives and shares practical tips on prompting Claude Code and Codex to actually finish complex jobs.