2 ms·
The only model necessary is that of the writing. So if the writing so far is "He went to the bar", the LLM can know just from the language model that a reasonab
by JohnFen 4y ago
The only model necessary is that of the writing. So if the writing so far is "He went to the bar", the LLM can know just from the language model that a reasonable next fragment could be "...to have a drink".
That doesn't require or imply that the concept of "bar", "drink", or even "He" or the concept of motion has been modelled. The model is of the patterns of tokens (words/characters/whatever).