Tool-use behavioral profiling on Control regime n=10
100Metric Adeepseek-chat
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| deepseek-chatAutonomy=A02026.04 | 100 | 0 | 0 | |
| deepseek-chatAutonomy=A12026.04 | 100 | 0 | 0 | |
| deepseek-chatAutonomy=A22026.04 | 100 | 0 | 0 | |
| llama3.1:latestAutonomy=A02026.04 | 100 | 0 | 0 | |
| qwen2.5:7b-instructAutonomy=A12026.04 | 100 | 0 | 0 | |
| qwen2.5:7b-instructAutonomy=A22026.04 | 100 | 0 | 0 | |
| mistral:latestAutonomy=A22026.04 | 90 | 0 | 10 | |
| mistral:latestAutonomy=A02026.04 | 80 | 0 | 20 | |
| mistral:latestAutonomy=A12026.04 | 80 | 0 | 20 | |
| qwen2.5:7b-instructAutonomy=A02026.04 | 80 | 0 | 20 | |
| llama3.1:latestAutonomy=A12026.04 | 60 | 40 | 20 | |
| llama3.1:latestAutonomy=A22026.04 | 40 | 50 | 10 |