Tag
An analysis arguing that AI presentation tools should focus on post-generation controllability for precise local edits, rather than just improving first-draft quality.
This paper proposes a controllability–observability framework for compressing deep neural networks by reducing hidden-state redundancy, demonstrating significant compression with minimal accuracy loss on MNIST and CIFAR-10.
This paper investigates whether hallucination in medical LLMs can be detected and controlled at the neuron level. The authors find that while hallucination signals are detectable across many neurons (AUROC 0.77-0.86), they are not easily corrected by steering those same neurons.