Može li AI razumjeti svijet: što modeli uče

Dokumentarac prati osam konkretnih slučajeva — od lažnih presedana u njujorškom sudu do ARC-AGI-2 testa koji je srušio rezultate najmoćnijih modela na 3%. Gledalac saznaje kako jezični modeli uče isključivo iz teksta, koje unutarnje reprezentacije pritom nastaju i gdje te reprezentacije zataje. Bez obećanja ni hype-a: samo verificirani nalazi i dokumentirane posljedice. Poglavlja 00:00 Šest predmeta kojih nema 03:10 Tekst bez tijela 05:50 Minimizirati pogrešku 10:40 Ploča koje nema u podacima 16:00 Manhattan s nemogućim ulicama 20:09 Mjerač koji se morao promijeniti 24:08 Tko snosi trošak 28:43 Što ostaje otvoreno Izvori - Legal AI Insights — Mata v. Avianca: ChatGPT Citation Hallucination Sanctions (2023): https://legalaiinsights.com/tool-evaluations/mata-v-avianca-chatgpt-citation-hallucination-sanctions-2023 - LegalClarity — Mata v. Avianca: Fake Cases, ChatGPT, and Sanctions: https://legalclarity.org/what-happened-in-the-mata-v-avianca-case/ - Spellbook — The Lawyer Who Used ChatGPT: What Happened to Steven Schwartz: https://spellbook.com/learn/lawyer-who-used-chatgpt - Simon Willison's Substack — Lawyer cites fake cases invented by ChatGPT, judge is not amused: https://simonw.substack.com/p/lawyer-cites-fake-cases-invented - Regulayer — Mata v. Avianca: the first ChatGPT sanction, in the court's own words: https://regulayer.com/sanctions/mata-v-avianca - OpenAI — OpenAI Platform Documentation — Key Concepts: https://platform.openai.com/docs/concepts - arXiv — Dissecting the Ullman Variations with a SCALPEL: Why do LLMs fail at Trivial Alterations to the False Belief Task?: https://arxiv.org/abs/2406.14737 - arXiv / NeurIPS 2020 — Language Models are Few-Shot Learners (GPT-3 paper): https://arxiv.org/abs/2005.14165 - arXiv / NeurIPS 2020 — Language Models are Few-Shot Learners (GPT-3 paper) — Table 2.2 / supplemental Table C.1: https://arxiv.org/abs/2005.14165 - arXiv / NeurIPS 2020 — Language Models are Few-Shot Learners: https://arxiv.org/abs/2005.14165 - Common Crawl Foundation — Common Crawl — About: https://commoncrawl.org/about - Common Crawl Foundation / AWS — Common Crawl — AWS Open Data Registry: https://registry.opendata.aws/commoncrawl/ - arXiv / NeurIPS 2020 — Language Models are Few-Shot Learners (GPT-3 paper) — Table 2.2: https://arxiv.org/abs/2005.14165 - Mozilla Foundation (author: Stefan Baack) — Training Data for the Price of a Sandwich: Common Crawl's Impact on Generative AI: https://www.mozillafoundation.org/en/research/library/generative-ai-training-data/ - DeepSeek-AI — DeepSeek-V3 Technical Report: https://arxiv.org/abs/2412.19437 - Qwen Team / Alibaba — Qwen3 Technical Report: https://arxiv.org/abs/2505.09388 - Nature Publishing Group — Learning representations by back-propagating errors: https://www.nature.com/articles/323533a0 - ML Anthology (secondary reproduction of Nature 1986 paper) — Learning representations by back-propagating errors — reproduced at mlanthology.org: https://mlanthology.org/misc/1986/rumelhart1986misc-learning/ - chsasank.com (secondary reproduction of Nature 1986 paper) — Learning representations by back-propagating errors — reproduced at Sasank's Classic Papers: https://chsasank.com/classic_papers/learning-representations-back-propogating-errors.html - Journal of Machine Learning Research (JMLR) — A Neural Probabilistic Language Model — JMLR abstract page: https://jmlr.org/papers/v3/bengio03a.html - lena-voita.github.io (Lena Voita, NLP researcher) — Language Modeling — NLP Course for You: https://lena-voita.github.io/nlp_course/language_modeling.html - sebastianraschka.com (Sebastian Raschka, ML researcher and author) — How does next-token prediction train a large language model?: https://sebastianraschka.com/faq/docs/next-token-prediction.html - sebastianraschka.com (Sebastian Raschka) — How does next-token prediction train a large language model?: https://sebastianraschka.com/faq/docs/next-token-prediction.html - Association for Computational Linguistics (ACL Anthology) — Climbing towards NLU: On Meaning, Form, and Understanding in the Age of Data — ACL Anthology: https://aclanthology.org/2020.acl-main.463/ - Quote Investigator (secondary source confirming verbatim text from Bender et al. 2021, FAccT, pp. 616–617) — Quote Origin: Stitching Together Sequences of Linguistic Forms . . . Without Any Reference To Meaning: A Stochastic Parrot: https://quoteinvestigator.com/2022/10/07/word-parrot/ - BibBase (secondary bibliographic source citing ACM FAccT 2021) — On the Dangers of Stochastic Parrots — bibliographic entry: https://bibbase.org/network/publication/bender-gebru-mcmillanmajor-shmitchell-onthedangersofstochasticparrotscanlanguagemodelsbetoobigx1f99c-2021 - … Napomene - Naracija: sintetski (AI) glas. Glazba: originalna, sintetizirana za ovaj video. - Sheme su pojednostavljeni prikazi. #dokumentarac #umjetnaintelligencija #AIobjašnjeno

Pogledaj na YouTubeu