Why can (a + b) + c differ from a + (b + c) for finite floating-point values?

Why can (a + b) + c differ from a + (b + c) for finite floating-point values?

1. Each grouping can round a different intermediate result.
2. Floating-point addition is never commutative.
3. Parentheses convert the operands to decimal.
4. Only negative values can lose precision.