List your company

Figurative-to-Literal Steering on Human Evaluation (sample of 100)

75Successful Sentences CountGPT-OSS-20B
37.5647.285766.72Apr 20, 2026

Evaluation results

MethodLinks
GPT-OSS-20BSteering status=Steer2026.0475
Llama-3.1-8BSteering status=Steer2026.0471
Qwen3-8BSteering status=Steer2026.0468
Gemma2-9BSteering status=Steer2026.0467
Gemma2-9BSteering status=Unsteer2026.0452
Qwen3-8BSteering status=Unsteer2026.0445
Llama-3.1-8BSteering status=Unsteer2026.0442
GPT-OSS-20BSteering status=Unsteer2026.0439