[cocogoatmain]

Provided you had the GPU compute to do so you could train the model to have less refusals, e.g. https://arxiv.org/abs/2407.01376

Quality of response/model performance may change though

There’s also nous research’s Hermes’ series of models, but those are trained on llama3.3 architecture and considered outdated now