I’ve only just started learning, so please let me know if there are any errors.

ML, learning in the broad sense

A series of processes for adjusting a model’s weight values

Learning in LLMs

The model has learned a “method” for converting relationships among words or the meaning of sentences into vectors

DAE (Denosing AutoEncoder)

1. The transformer embeds sentences, converts them into vectors, and learns the relationships among words.
2. It then learns the contextual meaning of each word and improves its ability to vectorize the meaning of sentences through masking and reconstruction.
3. The trained model is converted into a sentence embedding vector.

-> This is according to ChatGPT, so cross-validation is needed. -> It seems that the vectorization process and the denoising process are separate.