#29 Bálint Forgács: language, cognition, and AI скачать в хорошем качестве

#29 Bálint Forgács: language, cognition, and AI

A deep dive into how babies learn language with cognitive scientist Bálint Forgács from the ELTE BabyLab in Budapest. We explore the connectionist-vs-symbolic debate through the lens of infant cognition — how babies learn from one example rather than big data, why categories may be neither in the world nor in the mind but "transjective," and what newborns already know about speech before hearing a single word. In the second half, we turn to AI: why LLMs smuggle in symbols while pretending to be statistical machines, how they hack into our meaning-attribution system, and Sperber's Guru Effect as the perfect metaphor for why we find them so convincing. 00:00:00 Intro: the ELTE BabyLab 00:04:52 Bálint Forgács: metaphor, figurative meaning, and social cognition in babies 00:09:34 How do babies invent or discover structure? Has connectionism won? 00:16:58 Natural pedagogy: how babies learn from one example, not big data 00:22:17 Categories: the million-dollar question. Could it just be language? 00:28:33 The transjective view: categories are neither in the data nor in the mind 00:32:05 Newborns already distinguish speech from non-speech. Chimps don't. 00:46:04 Csibra's symbol sensitivity and Fodor's language of thought 00:48:52 Going from features to symbols is a huge problem (Gary Marcus) 00:53:23 LLMs pretend to be statistical machines, but they smuggle in symbols 00:56:32 Is the brain a computer or a statistics engine? The DNA argument 01:05:48 Categorical perception: why you can't hear French vowels after age 8 01:07:35 Newborn surprise: day-old babies detect conversational Q&A structure 01:13:02 Why is language hierarchical and not a zip file? 01:22:49 AI: why is inference not deterministic? Probabilistic but not statistical 01:29:14 LLMs hack into our attribution system 01:37:23 AI is brewing a problem in science 01:43:27 The Guru Effect: LLMs as the new Parisian philosophers 01:47:39 No intention, no real meaning. Chomsky's colorless green ideas. 01:50:25 Closing: "It can merge in two directions" I, scientist blog: https://balazskegl.substack.com Twitter:   / balazskegl   Artwork: DALL-E Music: Bea Palya    / @beapalyaofficial