That's not the case. They're saying you can find a collision easily (i.e. I can easily give you two files that have the same md5sum).
However, "crack an md5 hash" isn't what that means. First, you can't really "crack" a hash (there are infinitely many inputs that have the same hash), but even just going from a hash to any input with that hash is much harder.
Yeah, this is an important point. The known published attacks on MD5 are pretty narrow, so there are many circumstances where you can rely on MD5 hashes if you have to.
Was said a bit in jest, but in seriousness - only if you publish the hash and dox yourself, or the attacker gets your hash from a DB, some identifier, and finds you. That was the joke - in this case the comment does this. (I have no idea if or how much doxxed I didn't check)
Given I generated the hash with something like "head -c 1000 /dev/urandom | md5sum", I also don't really think a wrench attack would work here, even if you could find me. I don't know any input that produces that md5sum, no matter how many times you hit me with a wrench.
However, "crack an md5 hash" isn't what that means. First, you can't really "crack" a hash (there are infinitely many inputs that have the same hash), but even just going from a hash to any input with that hash is much harder.
That's a preimage attack, and a preimage attack for md5, according to wikipedia, remains theoretical https://en.wikipedia.org/wiki/MD5#Preimage_vulnerability.
I welcome anyone to give a counter example by giving me something that has an md5 hash of 08fc873f2aac5acce46ed751613472fe