1,286,497 views
Subscribe to the authors of the video on Telegram so as not to miss new materials: Pavel Komarovsky https://t.me/RationalAnswer and Igor Kotenkov https://t.me/seeallochnaya The following videos from the series about AI: - What new did GPT-4 bring - • GPT-4: What the new neural network has learned - About the confrontation between humanity and AI - • ChatGPT: Humanity versus the arts... Text version of the material about the evolution of language models: https://vc.ru/future/623774 Lecture by Igor Kotenkov about the principles of operation of transformers: • Transformer, explained in detail Ig... Listen to the episode as an audio podcast: - SoundCloud — / 94-how-chatgpt-works - Yandex Music — https://music.yandex.ru/album/23836417 - Other podcast aggregators — https://pod.link/1457434451/episode/1... Support the RationalAnswer project: - Patreon (in currency) – / rationalanswer - Boosty (in rubles) – https://boosty.to/RationalAnswer CONTENTS: 00:00 – Intro 00:31 – T9: language model in the phone 02:43 – Where do neural networks get word probabilities? 05:41 – Why language models can be creative 08:16 – 2018: GPT-1 and the Transformer architecture 11:47 – 2019: GPT-2, or 7000 Shakespeares in a neural network 15:25 – How the complexity and size of models are measured 22:07 – 2020: GPT-3, or the Incredible Hulk 26:13 – Prompts, or how to properly persuade a model 28:06 – January 2022: InstructGPT, or raising a shrew 33:51 – November 2022: ChatGPT – everyone is hyping it up! 38:05 – Let's sum it up