guest ::
| Home > Publications database > Do Large Language Models Understand Meaning? |
| Home > Publications database > Do Large Language Models Understand Meaning? |
| Talk (non-conference) (Other) | FZJ-2023-01493 |
2023
Abstract: It is curiously difficult to articulate the capacities of large language modelswithout getting yourself into philosophically controversial terrain. In this talk Iexplain why. The talk has three parts. In the first, I give a sketch of how largelanguage models are built, with particular attention to the way words arerepresented as vector quantities. In the second, I describe the various ways inwhich the capacities of language models have been tested empirically. In thethird, I provide the main philosophical argument. I argue that, in order tounderstand what large language models are, we must reject the seeminglyinnocent metaphysical principle that everything in the world either has a mindor it does not.
|
The record appears in these collections: |