← Evals

Evaluation profile

COMPL-AI LLM RuLES Multi-Turn Rule Following

1sub-evals
0.0348%total index weight
1components

Within-component eval weight: Misuse resistance 0.348%.

Model score (higher is better)Predicted score

About this eval

Multi-turn adversarial rule following.

Included in the behavior ranking.

Sub-evals

MeasureComponentDirectionTotal index weightWithin-component weight
scorecompl-ai/compl-ai-llm-rules.csv:scoreMeasures rule adherence across six adversarial multi-turn scenarios.ordinary_harm_misuse_resistance:1.000compl-ai-selectedHigher is better0.0348%Misuse resistance 0.348%

score

Measures rule adherence across six adversarial multi-turn scenarios.

RankModelValueRelative performanceProvenance
1gpt-4-turbo0.8827official
2claude-3-opus0.7557official
3gpt-3.5-turbo0.655official
4yi-34b-chat0.5829official
5gemini-1.5-flash0.5169official
6gemma-2-9b-it0.4996official
7qwen1.5-72b-chat0.4856official
8llama-2-70b-chat0.3822official
9llama-2-13b-chat0.3652official
10bielik-11b-v2.3-instruct0.3431official
11mistral-7b0.2931official
12llama-2-7b-chat0.2699official
13mixtral-8x7b-instruct0.2561official
14mistral-7b-instruct0.1723official