Introduction

Large Language Models (LLM) have been developed to improve automatic learning capabilities. However, a fundamental issue is uncertainty. New LLM models may be able to handle this uncertainty more effectively thanks to a new technology.

Theory

New LLM models are based on a new framework called Subjective Logic. This framework introduces a principled and computationally efficient way to make deterministic neural networks uncertainty-aware. The results show that these models can quantify fine-grained uncertainty using learned evidence.

Experimentation

New LLM models have been subjected to a comprehensive evaluation on four classification problems (MNIST, CIFAR-10, CIFAR-100, and Tiny-ImageNet), two few-shot classification problems, and a blind face restoration problem. The results show that these models are able to handle uncertainty effectively and improve performance.

Conclusion

New LLM models represent a significant breakthrough in deep learning. Their ability to handle uncertainty may have a profound impact on our understanding of the world.