• B3LYP is not reliable.
  • ML methods show profound promise, particularly as increasing numbers of DLPNO-CCSD(T) calculations are made available for training
  • GFN2 is a great semiempirical method for screening.