I work at a company that focuses on automating the data labelling process for computer vision. It is clear that generating massive amounts of labels, either by hand or automatically, without the ability to ensure a consistent level of quality across the dataset, is a problem. Which is why we are investing in automating the QA process for training data so mistakes like these don't happen:

https://blog.encord.com/post/automating-the-assessment-of-tr...

That's interesting, I took the opposite approach when realizing how bad label quality can be and built a site where everything is manually checked by moderators.

Most likely they trade out of the USDT position shortly after creating it which would minimise their exposure.

I don't think this would be the case, since these firms are actively market-making using USDT. I.e., they're not just turning fiat into crypto using USDT, but they are also market-making in USDT/BTC, USDT/ETH, etc.

My understanding is that Alameda is trading with fairly high frequency, so they would keep funds in USDT in order to facilitate those trades. Further, the on-chain data suggests that they don't trade out of USDT to USD – otherwise we'd see high corresponding USDT burns.