This paper examines the relation between the nature of updates in training curricula and graduates' skills and wages. Using vocational education and training (VET) curriculum texts as data, we apply natural language processing methods to identify the nature of changes in curriculum updates based on before-after comparisons of the skills the curricula contain. We introduce and measure two dimensions of curriculum updates: the 'novelty rate' (degree of new skills entering an updated curriculum) and the 'removal rate' (degree of old skills dropped from an old curriculum). By matching these dimensions to labour market data for VET graduates, we empirically investigate the relation between different types of curriculum updates and the wages of graduates with updated curricula compared to those of graduates with old curricula. We find non-linear relations and complementarities between the dimensions of updates and graduates' wages: while adding lots of new skills without removing old ones is not favourable, removing old skills is favourable across all curriculum updates. A further analysis on the actual skills that are added or removed illustrates the trade-offs that curriculum designers face. Given the economic importance of optimally preparing young workers for the labour market, our results provide crucial insights for educational policymakers.