Well, because it uses a content-addressed hash, you don't need to manage .torrent files or anything.
Also, if people make two torrents with the same files, and you download one of them, it only downloads from people who seed that torrent, not from people seeding the other torrent with the same files. Since IPFS has content addressed hashes each file, you can download from anyone who has that file.
If you read the rest of it, you'd realize your statement is only half of it. Magnet links only give you the torrent and tracker info, not a way to de-dupe or leverage a broader swarm based on availability of the files vs. a specific torrent.
Its quite a bit more, but I wouldn't blame you for making the comparison.
Side note: unless you're a dev really trying to get in on bleeding edge stuff, ignore the hype. The IPFS team is doing great, but there's a lot of work left to do.
The torrent TLDR is; theres a lot of quality-of-life things missing from torrents (if they were going to be loaded like webpages) and it turns out fixing some of those quality of life things requires solving some pretty hard technical challenges, even if they don't feel much different to the end user. Things like decentelized discovery, content address hashing (best thing since sliced bread), merkel trees for de-duplication, and change detection.
IPFS is much easier for someone to block and moderate as it isn't designed to be all that censorship resistant.
If you are looking for some "after market" content IPFS isn't what you want. If you are looking for something that could been the future of CDNs then then maybe it is for you.