From Likelihood to Fitness: Improving Variant Effect Prediction in Protein and Genome Language Models | Read Paper on Bytez