We thank all the reviewers for their valuable comments and suggestions. We will revise accordingly in the final version.$ Response to Reviewer_1: Notation: Here [d] is the set {1,2,\ldots,d} and $\psi_2$ is an Orlicz norm, which is also known as the variance proxy or sub-Gaussian norm. We will clearly define them in the revision to avoid any confusion. Proof: Thanks for your suggestion on making the proof section more sketched. We will highlight the high level idea of the proof to make it easier to follow. Experiments: We also have the numerical results that verify the theoretical results in Section 4.2 for stochastic block models. We will add them into the appendix in the revision. Best hierarchy: To be rigorous, SoS is the tightest *existing* convex relaxation hierarchy that is known to the research community [3]. This makes it a nature object for studying computational-statistical tradeoffs in the present moment. It is entirely possible that in the future even more powerful hierarchies are discovered, which could then become the next objects of interest for studying computational-statistical tradeoffs. We will make this clearer with a remark in the revision. Response to Reviewer_2: Title: Thanks for your suggestions. In this paper, we focus on SoS relaxation (the tightest existing relaxation), and use two major problems in statistical learning to demonstrate the fundamental limits of SoS relaxation. We will change our title to be more specific in the revision. Main message: We believe there is a misunderstanding in your summary of our main message. In fact, we show that the information-theoretical lower bound is *not* attainable by powerful convex relaxations, which demonstrates the gap between the information-theoretic (conceptual) and computationally-tractable (practical) optimality. Response to Reviewer_4: Thank you for the supportive comments.