Yann LeCun про ограничения всех LLM. Чуда AGI не будет. (35 мин, вместо 3 часов)

🌟 An interview with Yann LeCun, Chief Scientist at Meta and a Turing Award winner, reveals the limitations of modern language models and his vision for advancing artificial intelligence through the creation of systems capable of understanding the physical world and planning actions. I write helpful articles on my Telegram blog: https://t.me/blog_hanzo Links to the original video in the first comment 👇👇👇 👨🏻‍💻 Who am I? I create automation solutions for businesses through chatbots in Telegram and WhatsApp. I actively use neural networks in my projects. My journey includes 10 years of engineering experience, an MBA, process management at an IKEA plant, and three years of business analysis at 15 companies. Key Points The Limitations of Current Language Models in Understanding the Physical World The Difference Between Word Prediction and Real Context Understanding The JEPA Architecture: A New Approach to Representational Learning in AI The Importance of Self-Learning and the Impossibility of Achieving Intelligence through Language Alone Hierarchical Planning as a Key Component of Artificial Intelligence From Autoregressive Prediction to Energy-Based Models of Thinking Why Perfectly Unbiased AI is Impossible and Dangerous for Democracy Open Models as a Counterweight to the Monopoly on Information and Knowledge The Gradual Evolution of AI versus the Myth of "Sudden Superintelligence" The Future of Human-AI Interaction: Augmenting Capabilities, Not Replacing Them Discover how large language models and LLM agents work! Simple explanations, real-world examples, and their applications in education, medicine, finance, and more. A podcast for anyone curious about the future of artificial intelligence. Odkryj, jak działają duże modele językowe i agenci LLM! Proste wyjaśnienia, przykłady z życia wzięte i ich zastosowania w edukacji, medycynie, finansach i nie tylko. Podcast dla każdego, kto jest ciekawy przyszłości sztucznej inteligencji. Foundation of LLMs Reasoning Planning, tool use LLM agent infrastructure Retrieval-augmented generation Multimodal agents Human-agent interaction, personalization, alignment Multi-agent collaboration pretraining data (internet) tokenization neural network pretraining to post-training post-training data (conversations) hallucinations, tool use, knowledge, working memory #llm #chatgpt #neural networks #ai