Block-Element-Modifier names a component, its parts, and its variants in a way that's self-documenting just by reading the class name - no need to go hunting through a stylesheet to understand what .card__title--large means or where it lives.
.card { } /* Block */
.card__title { } /* Element */
.card--featured { } /* Modifier */
.card__title--large { }
Why it prevents specificity fights
Because every selector is a single flat class, BEM naturally avoids the deeply nested selectors that cause specificity wars in the first place. Overriding a style is almost always a matter of adding a modifier class, not fighting an ID selector three levels deep.
It reads verbosely at first. On any project with more than a handful of components and more than one developer touching the CSS, that verbosity pays for itself within a few weeks.
The naming convention is the whole point
BEM's block__element--modifier syntax looks verbose the first time you write it, and that's deliberate - the verbosity is what makes a class name unambiguous months later, in a file you didn't write, without opening the HTML to check what it's nested inside.
.card {}
.card__title {}
.card__title--large {}
.card--featured {}
What it actually solves
Specificity wars - the ones where someone reaches for !important because three nested selectors are all fighting over one element's color - mostly come from selectors that depend on DOM position rather than naming intent. Every BEM class is a single flat selector with equal specificity to every other BEM class, so the cascade stops being a factor in day-to-day styling. The tradeoff is longer class lists in the markup; for a team larger than one person, that trade is almost always worth making.