Back to explore

Model keputusan sebelum shell: AI yang tahu kapan tidak tahu

Sebelum agen coding AI menjalankan perintah ireversibel seperti rm -rf, model keputusan menilai keyakinan; keyakinan rendah berhenti dan meminta persetujuan manusia agar produksi tidak terhapus.

AI coding agent: "rm -rf $TARGET/" $TARGET = unset. One command can erase production. Instead, put a decision model before the shell. "irreversible: 0.33 confidence" Low confidence → STOP → human approval. The safety feature? AI that knows when it doesn’t know. #jev

· 2 likesOpen on X