Today's large language models (LLMs) routinely generate coherent, grammatical
and seemingly meaningful paragraphs of text. This achievement has led to
speculation that these networks are -- or will soon become -- "thinking
machines", capable of performing tasks that require abstract knowledge and
reasoning. Here, we review the capabilities of LLMs by considering their
performance on two different aspects of language use: 'formal linguistic
competence', which includes knowledge of rules and patterns of a given
language, and 'functional linguistic competence', a host of cognitive abilities
required for language understanding and use in the real world. Drawing on
evidence from cognitive neuroscience, we show that formal competence in humans
relies on specialized language processing mechanisms, whereas functional
competence recruits multiple extralinguistic capacities that comprise human
thought, such as formal reasoning, world knowledge, situation modeling, and
social cognition. In line with this distinction, LLMs show impressive (although
imperfect) performance on tasks requiring formal linguistic competence, but
fail on many tests requiring functional competence. Based on this evidence, we
argue that (1) contemporary LLMs should be taken seriously as models of formal
linguistic skills; (2) models that master real-life language use would need to
incorporate or develop not only a core language module, but also multiple
non-language-specific cognitive capacities required for modeling thought.
Overall, a distinction between formal and functional linguistic competence
helps clarify the discourse surrounding LLMs' potential and provides a path
toward building models that understand and use language in human-like ways.

今天的大型语言模型 (LLMs) 可以生成连贯的，符合语法的、有意义的文本段落，但在如人类思维一样的实际语言使用中，大多数测试需要功能语言能力，从认知神经科学的证据中，我们显示出 LLMs 显示出令人印象深刻（虽然不完美）的正式语言能力的任务，但在需要功能能力的许多测试中失败了。

大型语言模型中语言和思维的分离：认知角度

Dissociating language and thought in large language models: a cognitive perspective

The broader scope of our investigations is the search for the way in which
concepts and their combinations carry and influence meaning and what this
implies for human thought. More specifically, we examine the use of the
mathematical formalism of quantum mechanics as a modeling instrument and
propose a general mathematical modeling scheme for the combinations of
concepts. We point out that quantum mechanical principles, such as
superposition and interference, are at the origin of specific effects in
cognition related to concept combinations, such as the guppy effect and the
overextension and underextension of membership weights of items. We work out a
concrete quantum mechanical model for a large set of experimental data of
membership weights with overextension and underextension of items with respect
to the conjunction and disjunction of pairs of concepts, and show that no
classical model is possible for these data. We put forward an explanation by
linking the presence of quantum aspects that model concept combinations to the
basic process of concept formation. We investigate the implications of our
quantum modeling scheme for the structure of human thought, and show the
presence of a two-layer structure consisting of a classical logical layer and a
quantum conceptual layer. We consider connections between our findings and
phenomena such as the disjunction effect and the conjunction fallacy in
decision theory, violations of the sure thing principle, and the Allais and
Elsberg paradoxes in economics.

研究了量子力学模型和数学模型在概念和它们的组合中的应用，提出了一种概念组合的数学建模方案，阐明了量子力学原理在认知中的作用及其与概念形成基本过程的联系，并调查了其对人类思维结构的影响及其与决策理论、经济学等领域中的若干现象的关系。