Astra confidently wrong in real-world test, changes caused performance regressions
cnakazawa · x · 2026-09-05
Developer cnakazawa reports a hands-on test of the new model Astra: it was confidently and completely wrong, and its proposed changes delivered no meaningful performance improvement while causing regressions. The episode highlights the new model's overconfident-but-unreliable behavior in engineering tasks.
More from Fun
- IIT Bombay mechanical grad: only non-CS among 18 Stripe interviews, lands AI startup role — dhruv2038 · 2026-09-05
- GPT progression: GPT-3 text, GPT-4 images, GPT-5 video, GPT-6 3D? — PaulYacoubian · 2026-09-05
- When your advisor walks up mid-poster-grilling: the classic academic moment — docmilanfar · 2026-09-05
- Viral take: advisors and part-time VCs running on 10% effort are chasing fake relevance — vaibhavbetter · 2026-09-05
- GPT-6 Astra builds playable virtual piano with all six Brandenburg Concertos in ~11 minutes — DeryaTR_ · 2026-09-05
- GPT-6 hyped as AGI can't even render a snake with its head attached — SnooCheesecakes1893 · 2026-09-05