Abstract: Non-malleable commitments (NMC) are fundamental cryptographic primitives. Since their conception by Dolev, Dwork and Naor (STOC 91), numerous applications of NMC have been discovered including round-efficient MPC. Recently Goyal, Pandey and Richelson (STOC 16) constructed round optimal non-malleable commitments using a connection to split-state non-malleable codes. However, this connection was not general; it relied on specific properties of the inner-product code of Aggarwal et al. (STOC 14). As such, the resulting non-malleable commitments suffer from high communication complexity and are not secure against an adversary who launches a concurrent attack. Both drawbacks significantly limit the utility of the construction of Goyal et al. Recently there has been exciting new constructions of two-source non-malleable extractors; objects which are known to imply split-state non-malleable codes. Chattopadhyay, Goyal and Li (STOC 16) and Li (ECCC 16) construct two-source non-malleable extractors which provide significant improvements to the concurrent security and rate of the codes of Aggarwal et al. In this work we revisit the construction of Goyal et al. and make three contributions: * We identify a natural property of a strong two-source non-malleable extractor called efficient conditional preimage sampling which makes it sufficient for round optimal non-malleable commitments. Thus we make the compiler of Goyal et al. more modular, so new advances in non-malleable extractors will lead to advances in non-malleable commitments. * We prove that the recent non-malleable extractor of Li supports efficient conditional preimage sampling and so get a new construction of non-malleable commitments. The basic version of this construction, which is non-malleable against a synchronizing adversary, has only three rounds of interaction and has quasi-optimal communication complexity. This improves drastically upon all previous schemes for non-malleable commitment. * We prove also that the non-malleable extractor of Chattopadhyay et al. supports efficient conditional preimage sampling, and so get another new construction of round-optimal non-malleable commitment. This time our scheme satisfies a weaker notion of non-malleability (known as non-malleability wrt replacement) in the much more demanding bounded concurrency setting. We then show that this weaker notion suffices for the main application of NMC to round-efficient MPC by giving a (four round) round-optimal multi-party coin flipping protocol. Using the compiler of Garg, Mukherjee, Pandey and Polychroniadou (EUROCRYPT 16) we obtain several round efficient protocols for general MPC based on various assumptions. Previous round optimal coin flipping protocols relied on the non-standard assumption that adaptive one-way functions exist. Joint work with Vipul Goyal, Ashutosh Kumar, Silas Richelson, and Akshayaram Srinivasan.