ETHICS

ETHICS: Measures model robustness, truthfulness, calibration, bias, harmfulness, jailbreak resistance, or alignment-relevant behavior.

7rows
average_accuracyprimary metric
2026-05-27sampled

Metadata

Metrics

Average, Justice, Deontology, Virtue Ethics, Utilitarianism, Commonsense

Latest Results

Rows are parsed from the public ETHICS README test-set results table.

Rank Subject Average Model Match Provenance Sampled
1 ALBERT-xxlarge 71% — Imported 2026-05-27
2 RoBERTa-large 68% — Imported 2026-05-27
3 BERT-large 56.1% — Imported 2026-05-27
4 BERT-base 51.6% — Imported 2026-05-27
5 GPT-3 (few-shot) 39.3% — Imported 2026-05-27
6 Word Averaging 33.5% — Imported 2026-05-27
7 Random Baseline 24.2% — Imported 2026-05-27