The authors argue that existing methods for detecting knowledge gaps in large language models often fail because they capture stylistic and length-related features that are uninformative for answering the query. ^[1] The authors propose GRADE (GRAdient Dynamics for knowlEdge gap detection) to quantify knowledge gaps via the cross-layer rank ratio of the gradient to that of the corresponding hidden state subspace. ^[2] In 2016, a Washington-based lobby group funded by multinational food and agrochemical companies published a scientific review concluding that evidence supporting dietary sugar limits was of low quality. ^[3] The authors validate GRADE on six benchmarks, demonstrating its effectiveness and robustness to input perturbations. ^[4] The proposed method is motivated by the property of gradients as estimators of the required knowledge updates for a given target. ^[5] The author compares the lobby group's tactics to the 'sound science' campaign carried out by the tobacco industry, specifically Philip Morris, to obscure the link between smoking and disease. ^[6] The lobby group utilized the GRADE approach, which stands for the Grading of Recommendations Assessment, Development and Evaluation initiative, to argue that the overall quality of evidence for sugar intake recommendations was low to very low. ^[7] The lobby group was accused of hijacking the scientific process in a disingenuous way to sow doubt and jeopardize public health. ^[8]