Due to their high hardness, resistance to wear and excellent chemical stability, boride-based ceramic materials are important in machining processes such as cutting or grinding, but are also highly relevant to aerospace industry and geological exploration. Boron forms a wide variety of predominantly covalent-bonded, hence relatively brittle compounds with transition metals. In this talk, we will focus on transition metal diborides (TMBs) and study their strengthening and toughening mechanisms with the aid of ab inito molecular dynamics (AIMD). Namely, the homologous $\mathrm{TMB}_{2}$, where TM = Ti, Ta, Re, and W, will be discussed, starting with $\mathrm{TiB}_{2}$ which is already applied in industry. Then, we use the accurate data set produced during AIMD simulations to develop machine-learning interatomic potentials (MLIP). These allow us for running 'large'-scale molecular dynamics simulations that can uncover unknown phenomena as, e.g., stress-induced formation of extended defects that were not present in the smaller AIMD supercells. We will discuss trends in $\mathrm{TMB}_{2}$ mechanical properties considering the differences between AIMD and MLIP results – mainly due to size effects – as well as advantages and limitations of the two methods.