Literal-to-Figurative Steering on Human Evaluation (sample of 100)
15Successful SentencesGPT-OSS-20B
Evaluation results
| Method | Links | |
|---|---|---|
| GPT-OSS-20BSteering status=Steer2026.04 | 15 | |
| Llama-3.1-8BSteering status=Steer2026.04 | 12 | |
| Qwen3-8BSteering status=Steer2026.04 | 10 | |
| Gemma2-9BSteering status=Steer2026.04 | 9 | |
| Llama-3.1-8BSteering status=Unsteer2026.04 | 0 | |
| Qwen3-8BSteering status=Unsteer2026.04 | 0 | |
| Gemma2-9BSteering status=Unsteer2026.04 | 0 | |
| GPT-OSS-20BSteering status=Unsteer2026.04 | 0 |