Leave those texts out of the training data, and their sentences would be no less grammatical, but they would lose the ability to explain the grammar. Interestingly, the LLMs can synthesize these explanations, but only because they have “read” all the works of these adult linguists.

The child is not able to synthesize the explanations that the adult linguists have laboriously constructed. The child learns from examples, like the large language models, albeit with fewer examples.

They then cite the “consciously and laboriously … explicit version of the grammar” constructed by (adult, highly-educated) linguists as evidence that “child’s operating system is completely different from that of a machine learning program.” To me, it could be evidence to the contrary. They speak of the “miniscule data” and “minimal exposure to information” that lead a child to be able to distinguish between a grammatically well-formed sentence and one that is not. The authors assert that, “the human mind is a surprisingly efficient and even elegant system that operates with small amounts of information.” I have studied (and taught) information theory, and any measure I can imagine for the information provided to human brain during its 20 or so years of development into an educated, rational being is not small. We take morally inconsistent positions or weasel our way out of taking a position at all. The authors continue, “These differences place significant limitations on what these programs can do, encoding them with ineradicable defects.” But the defects they cite, to me, strikingly resemble defects in humans. Could they teach us something about how humans reason and use language? I have yet to encounter any scientist, even experts in machine learning, who are not surprised by the astonishing linguistic capabilities of these LLMs. One of the principles of science is that when an experiment yields a surprising result, we should be reluctant to dismiss the experiment and stubbornly cling to our preconceptions. What they do not consider is what these AIs may be able to teach us about humanity.Ĭhomsky, et al., state, “we know from the science of linguistics and the philosophy of knowledge that they differ profoundly from how humans reason and use language.” Do we know that? They seem much more confident about the state of the “science of linguistics and the philosophy of knowledge” than I am. Noam Chomsky, Ian Roberts, and Jeffrey Watumull, in “ The False Promise of ChatGPT,” (New York Times, March 8, 2023), lament the sudden popularity of large language models (LLMs) like OpenAI’s ChapGPT, Google’s Bard, and Microsoft’s Sydney.

Microsoft to do calendar view