Text 1: Computer scientist Yu argues that large language models "understand" the texts they generate. The models, she notes, successfully translate, summarize, and answer questions — tasks that seem to require comprehension. To insist that this is mere pattern matching, Yu contends, is to set a goalpost no system, human or machine, could ever clearly meet.
Text 2: Cognitive scientist Park finds Yu's argument too quick. The models, she argues, manipulate statistical regularities in token sequences without grounding them in any world. They produce fluent output about objects they have never perceived and propositions they cannot evaluate. Calling this "understanding," Park argues, conflates surface performance with cognitive substance.
Based on the texts, how would Park (Text 2) most likely respond to Yu's claim about understanding?
- A
She would agree that fluent output proves understanding.
- B
She would deny that language models can produce coherent text.
- C
She would propose that all software systems understand their outputs.
- Dcheck_circle
She would argue that performance can be impressive without constituting genuine understanding.
Explanation
Park distinguishes performance from grounded understanding. B captures her position. A contradicts her; C contradicts both authors; D is absurd.