Seems like the new Classifications feature uses GPT-3 text completion, and their similarity search model under the hood [0]: "The endpoint first searches over

Seems like the new Classifications feature uses GPT-3 text completion, and their similarity search model under the hood [0]: "The endpoint first searches over the labeled examples to select the ones most relevant for the particular query. Then, the relevant examples are combined with the query to construct a prompt to produce the final label via the completions endpoint." As a non-AI person, this sounds interesting. You wouldn't need to provide examples for every label you want, just enough that GPT-3 gets the idea. Is there prior art on this approach of text classification? [0] https://beta.openai.com/docs/api-reference/classifications https://beta.openai.com/docs/api-reference/classifications