{"id":"arxiv:2301.06627","title":"Dissociating language and thought in large language models","date":"2023-01-16","section":"Reasoning and the \"physics\" of language models","quality":0.052829,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3791d7ed420be517ded1ee7ae5558cf6afa948ea5b841a76ec3da8a9134ec565/","source_url":"https://arxiv.org/abs/2301.06627","text_available":true,"authors":["Kyle Mahowald","Anna A. Ivanova","Idan A. Blank","Nancy Kanwisher"],"outgoing":[{"source":"arxiv:2301.06627","target":"arxiv:2001.08361","evidence":"id","id":"arxiv:2001.08361","title":"Scaling Laws for Neural Language Models","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/82302458a6d03bc051b2e4fdb8277893a81d4bd6a029969045df6a8380904eac/"},{"source":"arxiv:2301.06627","target":"arxiv:2206.04615","evidence":"title","id":"arxiv:2206.04615","title":"Beyond the Imitation Game: Quantifying and extrapolating the capabilities of language models","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/70882486c1f8c46cb7b26cf100847f770eb7b39b69dee56b8a21141141ef814d/"}],"incoming":[{"source":"arxiv:2212.09196","target":"arxiv:2301.06627","evidence":"id","id":"arxiv:2212.09196","title":"Emergent Analogical Reasoning in Large Language Models","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/f74e529a278203e2c144ab7400a343836d9549cf1ae4843c584408448ad3f5a9/"},{"source":"arxiv:2407.15017","target":"arxiv:2301.06627","evidence":"title","id":"arxiv:2407.15017","title":"Knowledge Mechanisms in Large Language Models: A Survey and Perspective","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/8900dc2821aa61e59c5b1219cf9eb36dfaabf303c17d0064deaca653b5ec9a77/"},{"source":"doi:10.1038/s41586-024-07522-w","target":"arxiv:2301.06627","evidence":"title","id":"doi:10.1038/s41586-024-07522-w","title":"Language is primarily a tool for communication rather than thought","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/60cfed62cd01166a972691a7c0da286223d861d23feebc5fadd3fa446168a33c/"}],"text_parts":[{"start":0,"end":15659,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3791d7ed420be517ded1ee7ae5558cf6afa948ea5b841a76ec3da8a9134ec565/part-1.html"},{"start":15659,"end":31099,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3791d7ed420be517ded1ee7ae5558cf6afa948ea5b841a76ec3da8a9134ec565/part-2.html"},{"start":31099,"end":46720,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3791d7ed420be517ded1ee7ae5558cf6afa948ea5b841a76ec3da8a9134ec565/part-3.html"},{"start":46720,"end":61867,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3791d7ed420be517ded1ee7ae5558cf6afa948ea5b841a76ec3da8a9134ec565/part-4.html"},{"start":61867,"end":72564,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3791d7ed420be517ded1ee7ae5558cf6afa948ea5b841a76ec3da8a9134ec565/part-5.html"}],"contents":[{"title":"Dissociating language and thought in large language models","offset":0,"level":1,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3791d7ed420be517ded1ee7ae5558cf6afa948ea5b841a76ec3da8a9134ec565/part-1.html#text"},{"title":"Abstract","offset":62,"level":6,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3791d7ed420be517ded1ee7ae5558cf6afa948ea5b841a76ec3da8a9134ec565/part-1.html#text"},{"title":"The language-thought conflation","offset":1274,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3791d7ed420be517ded1ee7ae5558cf6afa948ea5b841a76ec3da8a9134ec565/part-1.html#text"},{"title":"Formal vs. functional linguistic competence","offset":8221,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3791d7ed420be517ded1ee7ae5558cf6afa948ea5b841a76ec3da8a9134ec565/part-1.html#text"},{"title":"What does linguistic competence entail?","offset":8269,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3791d7ed420be517ded1ee7ae5558cf6afa948ea5b841a76ec3da8a9134ec565/part-1.html#text"},{"title":"Formal linguistic competence","offset":8314,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3791d7ed420be517ded1ee7ae5558cf6afa948ea5b841a76ec3da8a9134ec565/part-1.html#text"},{"title":"Functional linguistic competence","offset":10201,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3791d7ed420be517ded1ee7ae5558cf6afa948ea5b841a76ec3da8a9134ec565/part-1.html#text"},{"title":"Motivation for the distinction between formal vs. functional linguistic competence","offset":11279,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3791d7ed420be517ded1ee7ae5558cf6afa948ea5b841a76ec3da8a9134ec565/part-1.html#text"},{"title":"The language network supports language processing in the human brain","offset":11762,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3791d7ed420be517ded1ee7ae5558cf6afa948ea5b841a76ec3da8a9134ec565/part-1.html#text"},{"title":"The language network does not support non-linguistic cognition","offset":12625,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3791d7ed420be517ded1ee7ae5558cf6afa948ea5b841a76ec3da8a9134ec565/part-1.html#text"},{"title":"LLMs have largely mastered formal linguistic competence in English","offset":15588,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3791d7ed420be517ded1ee7ae5558cf6afa948ea5b841a76ec3da8a9134ec565/part-1.html#text"},{"title":"Statistical language models: some fundamentals","offset":17085,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3791d7ed420be517ded1ee7ae5558cf6afa948ea5b841a76ec3da8a9134ec565/part-2.html#text"},{"title":"Large language models learn core aspects of human language processing","offset":23781,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3791d7ed420be517ded1ee7ae5558cf6afa948ea5b841a76ec3da8a9134ec565/part-2.html#text"},{"title":"LLMs perform well on benchmarks of diverse linguistic phenomena","offset":24802,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3791d7ed420be517ded1ee7ae5558cf6afa948ea5b841a76ec3da8a9134ec565/part-2.html#text"},{"title":"LLMs learn hierarchical structure","offset":26231,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3791d7ed420be517ded1ee7ae5558cf6afa948ea5b841a76ec3da8a9134ec565/part-2.html#text"},{"title":"LLMs learn linguistic abstractions","offset":28358,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3791d7ed420be517ded1ee7ae5558cf6afa948ea5b841a76ec3da8a9134ec565/part-2.html#text"},{"title":"LLMs learn constructions","offset":31961,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3791d7ed420be517ded1ee7ae5558cf6afa948ea5b841a76ec3da8a9134ec565/part-3.html#text"},{"title":"LLMs are predictive of activity in the human language network","offset":33567,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3791d7ed420be517ded1ee7ae5558cf6afa948ea5b841a76ec3da8a9134ec565/part-3.html#text"},{"title":"Using LLMs as models of formal linguistic competence in humans","offset":36385,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3791d7ed420be517ded1ee7ae5558cf6afa948ea5b841a76ec3da8a9134ec565/part-3.html#text"},{"title":"Excessive reliance on statistical regularities","offset":37606,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3791d7ed420be517ded1ee7ae5558cf6afa948ea5b841a76ec3da8a9134ec565/part-3.html#text"},{"title":"Unrealistic amounts of training data","offset":37658,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3791d7ed420be517ded1ee7ae5558cf6afa948ea5b841a76ec3da8a9134ec565/part-3.html#text"},{"title":"Insufficient tests on languages other than English","offset":37700,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3791d7ed420be517ded1ee7ae5558cf6afa948ea5b841a76ec3da8a9134ec565/part-3.html#text"},{"title":"Non-augmented LLMs fall short on functional linguistic competence","offset":37756,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3791d7ed420be517ded1ee7ae5558cf6afa948ea5b841a76ec3da8a9134ec565/part-3.html#text"},{"title":"A. Fine-tuning on the task and the challenge of closed models","offset":39438,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3791d7ed420be517ded1ee7ae5558cf6afa948ea5b841a76ec3da8a9134ec565/part-3.html#text"},{"title":"B. Generalizable, robust performance","offset":39505,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3791d7ed420be517ded1ee7ae5558cf6afa948ea5b841a76ec3da8a9134ec565/part-3.html#text"},{"title":"Formal Reasoning","offset":39547,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3791d7ed420be517ded1ee7ae5558cf6afa948ea5b841a76ec3da8a9134ec565/part-3.html#text"},{"title":"World models 1: factual and commonsense knowledge","offset":42351,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3791d7ed420be517ded1ee7ae5558cf6afa948ea5b841a76ec3da8a9134ec565/part-3.html#text"},{"title":"World models 2: situation tracking","offset":46129,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3791d7ed420be517ded1ee7ae5558cf6afa948ea5b841a76ec3da8a9134ec565/part-3.html#text"},{"title":"Social reasoning","offset":48550,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3791d7ed420be517ded1ee7ae5558cf6afa948ea5b841a76ec3da8a9134ec565/part-4.html#text"},{"title":"Language input can bootstrap functional competence capabilities","offset":52800,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3791d7ed420be517ded1ee7ae5558cf6afa948ea5b841a76ec3da8a9134ec565/part-4.html#text"},{"title":"Toward models that use language like humans","offset":54307,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3791d7ed420be517ded1ee7ae5558cf6afa948ea5b841a76ec3da8a9134ec565/part-4.html#text"},{"title":"Concluding remarks","offset":60468,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3791d7ed420be517ded1ee7ae5558cf6afa948ea5b841a76ec3da8a9134ec565/part-4.html#text"},{"title":"Glossary","offset":63931,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3791d7ed420be517ded1ee7ae5558cf6afa948ea5b841a76ec3da8a9134ec565/part-5.html#text"},{"title":"Highlights","offset":67162,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3791d7ed420be517ded1ee7ae5558cf6afa948ea5b841a76ec3da8a9134ec565/part-5.html#text"},{"title":"Outstanding Questions","offset":68218,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3791d7ed420be517ded1ee7ae5558cf6afa948ea5b841a76ec3da8a9134ec565/part-5.html#text"},{"title":"Acknowledgements","offset":71271,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3791d7ed420be517ded1ee7ae5558cf6afa948ea5b841a76ec3da8a9134ec565/part-5.html#text"},{"title":"Funding Sources","offset":72091,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3791d7ed420be517ded1ee7ae5558cf6afa948ea5b841a76ec3da8a9134ec565/part-5.html#text"},{"title":"Conflicts of Interest","offset":72491,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3791d7ed420be517ded1ee7ae5558cf6afa948ea5b841a76ec3da8a9134ec565/part-5.html#text"}],"text_code_points":72564,"source_markdown":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3791d7ed420be517ded1ee7ae5558cf6afa948ea5b841a76ec3da8a9134ec565/source.md"}