Machine learning uses before the large language model hype also sometimes had some ethical problems – privacy of the data used to train them, accountability for the results/decisions coming out of them, … But at least it still was some kind of deliberate engineering, design decisions would be made about which data to feed into them, how to preprocess the input, how to configure what should come out of them, … This whole „We train a huge model on a huge pile of everything and then use it for everything afterwards.“ feels so … dirty.