Machine Learning APIs

src: (Chen et al., 2021)^chen2021did, OR

Machine Learning APIs are weird. Usually we deal with APIs because they are well-defined and predictable (and oftentimes deterministic, i.e. a pure function). Ignoring the actual API scaffolding here (the request/response schema), which I assume would follow API convention, the key output of the API, the predictions from the ML model, are incredibly