AI glasovni asistent govor: od podataka do rečenice

Dokumentarac prati kako jezični modeli poput GPT-3 i ChatGPT-a uče iz 300 milijardi tokena preuzetih s weba, uključujući autorski zaštićene knjige. Prikazuje mehanizam pažnje transformer arhitekture i RLHF postupak usklađivanja koji smanjuje halucinacije, a ne uklanja ih. Na tri dokumentirana slučaja – pravnom, zdravstvenom i financijskom – film pokazuje što se dogodi kada stroj generira uvjerljiv, ali netočan tekst. Poglavlja 00:00 Prizor bez presedana 02:20 Što stroj čita 06:22 Mehanizam pažnje 09:47 Povratna veza čovjeka 12:27 Kako se mjeri znanje stroja 15:34 Šest predmeta koji ne postoje 17:56 Tko snosi odgovornost 21:42 Tri nalaza i jedna granica Izvori - CourtDocket — Mata v. Avianca: Fake ChatGPT Cases, Sanctions, and Fallout: https://courtdocket.org/mata-v-avianca-fake-chatgpt-cases-sanctions-and-fallout/ - Spellbook — The Lawyer Who Used ChatGPT: What Happened to Steven Schwartz: https://spellbook.com/learn/lawyer-who-used-chatgpt - Aaron & Partners (UK law firm) — Mata v Avianca: The Hidden Dangers of Using AI for Legal Advice: https://www.aaronandpartners.com/news/mata-v-avianca-the-hidden-dangers-of-using-ai-for-legal-advice/ - Legal AI Governance — Mata v. Avianca, Inc.: https://legalaigovernance.com/tracker/cases/mata-v-avianca/ - 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 - Common Crawl Foundation — About – Common Crawl: https://commoncrawl.org/about - arXiv / NeurIPS 2020 — Language Models are Few-Shot Learners (arXiv:2005.14165): https://ar5iv.labs.arxiv.org/html/2005.14165 - arXiv / EleutherAI — The Pile: An 800GB Dataset of Diverse Text for Language Modeling (arXiv:2101.00027): https://ar5iv.labs.arxiv.org/html/2101.00027 - arXiv / EleutherAI — The Pile: An 800GB Dataset of Diverse Text for Language Modeling (arXiv:2101.00027): https://arxiv.org/abs/2101.00027 - TorrentFreak — Anti-Piracy Group Takes Prominent AI Training Dataset 'Books3' Offline: https://torrentfreak.com/anti-piracy-group-takes-prominent-ai-training-dataset-books3-offline-230816/ - CourtListener / PACER — Huckabee et al v. Meta Platforms, Inc. et al – CourtListener Docket: https://www.courtlistener.com/docket/67890942/huckabee-v-meta-platforms-inc/ - Variety — Sarah Silverman Sues Meta and OpenAI, Alleging Copyright Infringement: https://variety.com/2023/digital/news/sarah-silverman-sues-meta-openai-copyright-infringement-1235665185/ - AI Lawsuit Tracker — New York Times v. OpenAI & Microsoft – AI Lawsuit Tracker: https://ailawsuittracker.com/cases/new-york-times-v-openai/ - arXiv / Cornell University — arXiv:1706.03762 – Attention Is All You Need: https://arxiv.org/abs/1706.03762 - NeurIPS / Curran Associates — Attention is All you Need – NeurIPS Proceedings Abstract: https://proceedings.neurips.cc/paper_files/paper/2017/hash/3f5ee243547dee91fbd053c1c4a845aa-Abstract.html - arXiv / Cornell University — Attention Is All You Need – ar5iv HTML version: https://ar5iv.labs.arxiv.org/html/1706.03762 - OpenAI — OpenAI Blog – Improving language understanding with unsupervised learning: https://openai.com/index/language-unsupervised/ - OpenAI — Improving Language Understanding by Generative Pre-Training (GPT-1 paper): https://cdn.openai.com/research-covers/language-unsupervised/language_understanding_paper.pdf - Hugging Face (referencing Radford et al. 2019) — Language Models are Unsupervised Multitask Learners (GPT-2 paper overview): https://huggingface.co/transformers/v2.8.0/model_doc/gpt2.html - arXiv / Cornell University — arXiv:2005.14165 – Language Models are Few-Shot Learners: https://arxiv.org/abs/2005.14165 - Lambda Labs (Lambda AI) — OpenAI's GPT-3 Language Model: A Technical Overview – Lambda Labs Blog: https://lambda.ai/blog/demystifying-gpt-3 - NeurIPS Proceedings / Curran Associates — Training language models to follow instructions with human feedback (NeurIPS 2022 Abstract): https://proceedings.neurips.cc/paper_files/paper/2022/hash/b1efde53be364a73914f58805a001731-Abstract-Conference.html - NeurIPS Proceedings / Curran Associates — Training language models to follow instructions with human feedback — NeurIPS 2022 Proceedings: https://proceedings.neurips.cc/paper_files/paper/2022/hash/b1efde53be364a73914f58805a001731-Abstract-Conference.html - OpenAI / arXiv — Training language models to follow instructions with human feedback (arXiv:2203.02155) — ar5iv HTML: https://ar5iv.labs.arxiv.org/html/2203.02155 - … Napomene - Naracija: sintetski (AI) glas. Glazba: originalna, sintetizirana za ovaj video. - Sheme su pojednostavljeni prikazi. #dokumentarac #AI #jezičnimodeli

Pogledaj na YouTubeu