Persuading large language models to comply with objectionable requests.

We found that the treatment effect was smaller in magnitude but still highly significant ( B = 0.069; 95% CI [0.066, 0.072]; z = 42.21, P < 0.001, see SI Appendix , Table S4 Model 2 and SI Appendix , Table S5 for per principle models).

also in 132,142 other papers