2/ 해법은 더 큰 모델이 아니라 '문맥'이었습니다. 프로젝트 자체 규칙(RAG)을 주입하니 specificity가 23% → 82%(약 4배)로, 종합 balance는 2배로.
2트랙: Unity 6 지식 + 실제 버그 리뷰.
코드: https://t.co/VdDuH7mxf9
데이터: https://t.co/j98GAd3HNg
#Unity#LLM#게임개발
2/ The fix wasn't a bigger model — it was context.
2 tracks: Unity 6 knowledge + real bug review.
Code: https://t.co/VdDuH7mxf9
Data: https://t.co/j98GAd3HNg
#Unity#LLM#gamedev
1/ Can today's top code models find real bugs in real Unity 6 game code — without crying wolf?
I built UnityBench to measure it, on bugs reconstructed from my shipping game's version-control history.
Finding: every frontier model over-flags. 🧵