On Distinguishing Capability Elicitation from Capability Creation in Post-Training: A Free-Energy Perspective
Redefining the boundaries between model behavior refinement and genuine functional expansion.
Debates about large language model post-training often treat supervised fine-tuning (SFT) as imitation and reinforcement learning (RL) as discovery. But this distinction is too coarse. What matters is whether a training procedure increases the probability of behaviors the pretrai...
ArXiv CS.AI
Source
Generated insight by OnAirToday AI
This research provides a rigorous framework for determining whether post-training simply refines a model's existing knowledge or forces the development of fundamentally new cognitive abilities. Understanding this boundary is critical for accurately assessing the safety risks and functional limits of future generative AI systems.
Comments
Keep comments relevant and respectful. No links, advertising, self-promotion, or spam.