Hacking Large Language Models, Manifold Hypothesis - Dr. Michael Robinson | Unorthodox Views 010 скачать в хорошем качестве

Hacking Large Language Models, Manifold Hypothesis - Dr. Michael Robinson | Unorthodox Views 010

In this fascinating episode of Unorthodox Views, I sit down with Dr. Michael Robinson — mathematician, statistician, and professor at American University — to discuss two groundbreaking papers that challenge fundamental assumptions in large language models (LLMs). The two papers we discuss hare at arxiv: Michael Robinson, Sourya Dey, Tony Chiang "Token embeddings violate the manifold hypothesis" https://arxiv.org/abs/2504.01002 and Michael Robinson, Sourya Dey, Taisa Kushner "Probing the topology of the space of tokens with structured prompts" https://arxiv.org/abs/2503.15421 Please subscribe, like, and share if you want to see more deep-dive conversations with fascinating people! I am Dr. Krzysztof Pietroszek and you are watching Unorthodox Views podcast available on YouTube, Spotify and X. You can also support me on Patreon and follow me on X. Patreon:   / unorthodoxviews   My personal X profile: https://x.com/@holoprof Unorthodox Views YouTube playlist:    • Unorthodox Views 001 | Wave Function, Quan...   Unorthodox Views on X: x.com/@viewsunorthodox Unorthodox Views Spotify playlist: https://open.spotify.com/show/5yiTKGN... #LLM #LargeLanguageModels #AIresearch #TokenEmbeddings #ManifoldHypothesis #Singularities #LanguageModelHacking #PromptEngineering #MachineLearning #StatisticalModeling #GPT2 #LLaMA #MistralAI #Pythia #TokenSpace #EmbeddingSpace #