The paper discusses the challenges and opportunities presented by Big Data in modern society. It highlights the unique features of Big Data, such as massive sample size and high dimensionality, which introduce computational and statistical challenges. These challenges include noise accumulation, spurious correlations, incidental endogeneity, and measurement errors. The authors emphasize the need for new computational and statistical paradigms to address these issues. They provide an overview of the impact of these features on statistical inference and computational methods, and introduce several new perspectives on Big Data analysis. Key topics include the viability of sparsest solutions in high-confidence sets and the limitations of exogenous assumptions in statistical methods. The paper also discusses the implications of incidental endogeneity, which can lead to wrong statistical inferences and scientific conclusions. Finally, it reviews various statistical methods and computational techniques that have been developed to handle these challenges, including penalized quasi-likelihood, sparsest solution in high-confidence sets, and sure independence screening.The paper discusses the challenges and opportunities presented by Big Data in modern society. It highlights the unique features of Big Data, such as massive sample size and high dimensionality, which introduce computational and statistical challenges. These challenges include noise accumulation, spurious correlations, incidental endogeneity, and measurement errors. The authors emphasize the need for new computational and statistical paradigms to address these issues. They provide an overview of the impact of these features on statistical inference and computational methods, and introduce several new perspectives on Big Data analysis. Key topics include the viability of sparsest solutions in high-confidence sets and the limitations of exogenous assumptions in statistical methods. The paper also discusses the implications of incidental endogeneity, which can lead to wrong statistical inferences and scientific conclusions. Finally, it reviews various statistical methods and computational techniques that have been developed to handle these challenges, including penalized quasi-likelihood, sparsest solution in high-confidence sets, and sure independence screening.