The recent advancements in the field of explainability for machine learning models have seen a shift towards integrating high-dimensional data with deep generative models, addressing the gap between modern generative techniques and classical explainability methods. Innovations in probabilistic frameworks are enabling more rigorous and transparent communication of local example-based explanations, enhancing the quality of peer discussions and research. Additionally, there is a notable trend towards efficiency in generating uncertainty-aware explanations, with methods like Fast Calibrated Explanations offering significant speedups for real-time applications without compromising on uncertainty quantification. Furthermore, the integration of natural language explanations with clinical predictions, such as in EchoNarrator, is demonstrating the potential to increase trust and usability in medical AI applications. Lastly, the direct optimization of explanations for specific desirable properties is providing more control and consistency in the generation of explanations tailored to specific tasks.