The LLM is sampled to create just one-token continuation on the context. Presented a sequence of tokens, just one token is drawn in the distribution of possible upcoming tokens. This token is appended to your context, and the process is then recurring. Unsurprisingly, business enterprises tha