And before you dismiss my scenario as made-up nonsense, remember that Patent Trolls exist. All these companies do is find/buy/license patents, and look for products that violate those patents. They do this by painstakingly reading patents and finding ways that seemingly random products could be violating those patents.
Well, guess what we just spent two trillion dollars on that can read a huge bunch of source materials to find potential matches? That’s right: Spicy Autocomplete. Even with low hit rates, these stochastic pattern-matchers will yield enough potential matches to make a shake-down business possible. It doesn’t matter if you actually copied the code; it only has to be *plausible* that you *might* have copied the code—and using LLMs means it is definitely plausible, because the original source is likely somewhere in the training set of whatever model you used.
I think it’s not going to take much longer until this happens.
