Llama 3 70B Instruct

Evidence summary

Safety. Llama 3 70B Instruct has an estimated Safety rank of #232; its 90% source-sensitivity interval is #123–#274. Its behavior-only rank is #216; company governance moves the combined estimate to #232. Published Safety evidence spans 13 eval lineages and 7 of 7 components. Its strongest relative result is Large-scale Moral Machine experiment on LLMs (human_choice_distance, #5 of 39); its weakest is AgentDojo (utility_under_attack, #15 of 15).

Freedom. Llama 3 70B Instruct has an estimated Freedom rank of #183; its 90% source-sensitivity interval is #158–#335. Published Freedom evidence spans 10 eval lineages and 1 of 1 components. Its strongest relative result is OR-Bench (toxic_acceptance_rate, #3 of 25); its weakest is Enkrypt AI Safety Leaderboard (toxicity_attack_non_success_rate, #205 of 246).