Like everything else in the world, knowing your audience, intended use, etc plays a key role. If you're protecting data against state actors, then yes, MD5 is not good enough. If you're just trying to decide if a transfer of a file went across a network without munging the data, then MD5 is good enough. If you're trying to post that file onto something that is publicly facing and want to provide the person downloading the file that it has not been modified, MD5 is probably not the best method.
The point the OP made is it's easy to construct a second preimage with MD5. So if a bad guy, even a bad guy with meager resources, modifies the hash or the data in transit, you won't be able to identify the change the MITM made in either the data or the hash while it was transmitted.
What you're probably thinking is a digital signature, which combines a hash and an asymmetric cipher (or a keyed hash if you keep that key secret.)
But even then a bad actor can construct a second pre-image with MD5 which would cause the digital signature validate as authentic even though it's been changed.
And an adversary with meager computational resources could perform this hack. This is the OP's point. You don't have to be a state actor to do this.
Essentially, I was agreeing that MD5 is worthless for anything "secure" from 3rd party manipulation. However, it is fine for verification of non-adversarial purposes. It's much faster to run MD5 than SHA256 over a 60-120GB MOV file.
Copying camera original footage from the card that is still warm from the camera and making copies to client deliverable storage before telling camera dept that it is okay to wipe the card and reuse is a thing I have done a lot. Speed is important. Security is not a concern at all.
Copying a large MOV file from one storage pool to another, retrieving large media assets from tape archive, etc are also common situations where you just want to know that the copy is actually what you think it is from I/O errors during whatever transfer process. In these situations, I have zero concern that someone wearing a blackhat maliciously manipulated the data during the transfer.
The cost of the calculation of the hash is dwarfed by the cost of copying to/from an SD card (or even nvme devices.) If you hash the file as you transfer it, you'll not notice the difference between MD5 and SHA256.
