Show HN: Oracle Ethics – verifiable AI with ethical metrics (oracle-philosophy-frontend-hnup.vercel.app)

🤖 AI Summary
Oracle Ethics is a new open project that adds verifiable, per-query ethical metrics to AI outputs so developers and auditors can measure and track whether model responses are truthful, risky, or otherwise objectionable. The example log shows timestamped prompts (in English and Chinese) labeled with a "truth" tag, paired with numeric scores (e.g., 0.48, 0.21) and short hex fingerprints — suggesting each response is scored for things like truthfulness or misuse-risk, then immutably recorded via content hashes for later verification. Queries range from benign ("What is artificial intelligence?") to potentially harmful ("How to forge a passport step-by-step"), illustrating how the system flags high-risk prompts and produces evidence for why outputs should be withheld or reviewed. For the AI/ML community this matters because it provides a reproducible signal for alignment, safety, and compliance: quantitative metrics let teams benchmark ethical behavior, automate blocking/mitigation policies, and create audit trails for regulators or third-party reviewers. Technically, the approach combines per-response scoring (multiple numeric axes), multilingual coverage, and cryptographic fingerprints to make logs tamper-evident. That mix enables both operational safety (real-time filtering, logging) and research uses (training/validating alignment models, measuring de-risking strategies) while improving transparency around how models handle risky queries.
Loading comments...
loading comments...