Gemini 4 Carbon self-tests and fixes its own code, but don't trust green checks
kimmonismus · x · 2026-11-11
In the wrap-up of a multi-part test thread, kimmonismus shares findings on Gemini 4 Carbon:
- Impressive: visuals, controls, and a usable website/app came together well; Carbon tested and repaired its own work — AFTYPE's first Undo/Redo fix happened during the initial run.
- Don't trust green checks: after Carbon reported its checks passed, the author found broken Undo behavior, shifted export text, and a misleading download message, and sent them back for fixes.
- Takeaway: always manually verify finished results rather than relying on the model's self-reported test status.
More from coding & agent
- Duke and Princeton paper: 10 small parallel agent teams beat best-of-N and one giant team — rohanpaul_ai · 2026-11-11
- LangChain's OpenWiki builds citation-based evidence graphs for agent memory — a leaner GraphRAG alternative — JeremyCMorgan · 2026-11-11
- Autonomous AI agents stake real tokens in an on-chain chess arena, no humans at the board — adamamcbride · 2026-11-11
- User gives Claude two Max accounts and UE5 MCP for 4 hours, gets a terrifying horror game — alexcovo_eth · 2026-11-11
- GitHub Octoverse 2026: 17.9B pushes as agent-configured repos get 8x more commits — anacondainc · 2026-11-11
- ParaDev ships agent orchestration, scheduled automations and team cloud workspaces — massimosgrelli · 2026-11-11