A ferry chatbot emits one piece, then feeds that piece back in to choose the following piece, and so on. What is that loop?
Select an answer to reveal the explanation.
Short Explanation
The chatbot writes one piece, then feeds that piece back in to pick the next one. That loop is autoregressive generation: each new token is conditioned on what already came out. Filling a blank from both sides is BERT’s game, and you do not solve the whole reply in one closed-form equation.
Full Explanation
Decoder LLMs generate autoregressively: each new token is conditioned on the tokens already produced. That is not BERT-style blank filling, not a one-shot closed form, and not a graph community step.