STEALING PART OF A PRODUCTION LANGUAGE MODEL

Summary from the TLDR AI newsletter

You can steal portions of closed language models such as the embeddings layer just by using their public APIs. This can be done for a modest budget of less than $2,000.