If LLMs Can Decide Without Fine-Tuning, Do We Still Need Models Like Jev?
If the options have meaning, could a language model's learned probability distribution already contain the ability to choose? I tried reading decisions directly from existing models before deciding whether to add a speci…