ad_hominem

Pushshift is a single person with some very strong political opinions who has specifically used his datasets to attack political opponents. Frankly I wouldn't trust his data to be untainted.

These models really need to be trained on more official data sources, or at least something with some type of multi-party oversight rather than data that effectively fell off the back of a truck.

If some guy used his bottle as a hammer for several months before turning it in for recycling, then someone cuts their hand on a jagged edge or drinks a glass sliver, that's a multi-million dollar lawsuit. Unlikely maybe but at Coca-Cola scale I imagine things like that will happen eventually.

These companies would have to have sophisticated QA processes, or melt down and re-cast the bottles to ensure the food safety of their supply chain against millions of chaos monkeys.

Edit: Here I just did a Google search for the butthurt downvoters that proves there is extra QA that needs to be done. From an Oregon glass bottling reuse program:[1]

> Among other attributes, the machinery features an electronic sensor that uses X-ray equipment to image each bottle, detecting flaws in the glass, as well as mold and other contaminants, Bailey said. That step will reject any bottles that are chipped or contaminated.