Seonglae Choarxiv.org
https://arxiv.org/pdf/2505.11388
For black-box text-embedding-3-small embedding model
arxiv.org
https://arxiv.org/pdf/2408.00657v1
SAE Probing for label prediction.
While using only embeddings shows higher performance, obtaining interpretable NLP features is a notable advantage.
arxiv.org
https://arxiv.org/pdf/2502.04382
Token embedding → text embedding by pooling
www.arxiv.org
https://www.arxiv.org/pdf/2506.04373