API-Protected LLMs Leak Proprietary Details Through Logits
API-protected LLMs leak proprietary details through logits, a "back door" that reveals model training data & objective function. Researchers find API calls can extract full logit vector, compromising IP of LLM providers.
This is a Plain English Papers summary of a research paper called Logits of API-Protected LLMs Reveal Proprietary Model Details, Researchers Find. If you like these kinds of analysis, you should join AImodels.fyi or follow me on Twitter. Overview Large language models (LLMs) have become increasingly popular and powerful, but their inner workings are often opaque Researchers investigated whether the "logits" (outputs) of API-protected LLMs can reveal sensitive information about the model Plain English Explanation The paper examines whether the numerical outputs or "logits" f...