It seems to be a prototypical example of a problem that is very simple to state yet extremely hard to solve. In one view it represents a large class of similar problems, so progress on general methods for that class would be very significant.
What we know
It has been checked for enormous numbers, and in 2019 Terence Tao showed that almost all starting values get very small.
What blocks us
Mathematics may lack the tools; it is a simple rule with no known structure to exploit.
Current leads
Probabilistic methods, computer-assisted proofs.
Sub-questions
Can Tao's 'almost all' result be upgraded from logarithmic density to the usual natural density?
Do almost all starting numbers drop below one fixed bound? Tao says this may be almost as hard as the full conjecture.
Could some sequence grow forever, or could there be infinitely many loops?