Dev Notes GPT-6 Astra Still Needs Babysitting, Makes Wrong Calls on Physics Engine Changes
yacineMTB · x · 2026-09-05
Developer yacineMTB offers a counterpoint to the Astra hype: the model still needs babysitting, repeatedly making the wrong calls on changes to his physics engine. "Maybe GPT-7 is when I don't have to read the code anymore," he writes — a grounded first-hand check against the flood of one-shot demo clips.
Related event: Dev Finds GPT-6 Astra Still Needs Babysitting for Coding(2 posts)→
More from coding & agent
- seq-mcp: An MCP Server Giving AI Agents Controlled Read Access to Seq Logs — modelcontextprotocol · 2026-09-05
- Crew: An MIT-Licensed Desktop App Where Multiple LLM CLIs and People Share One Project — Classic-Grab-2866 · 2026-09-05
- Why agents name folders zz: reverse alphabetical tool ordering surfaces messages first — morqon · 2026-09-05
- After AI wrote his whole Python package, he asks: how can we ever trust AI code in production? — Lukee67 · 2026-09-05
- Dev: arbiter system wrote 85% of my code since Gemini 2.5 Pro — now it's down to 5% — Sauers_ · 2026-09-05
- Dev uses ChatGPT to resurrect shelved project with streaming and MiniMax H3 support — flngr · 2026-09-05