Exploring AI Learning, the specifics of Large Language Models, including architecture, prompt engineering, fine-tuning, RAG, and their application in text-based tasks.
Watch or Listen on YouTube Conditional Memory And DeepSeek Engram: When Lookup Beats More Compute Introduction Bigger models keep winning, but the reason is not always “more intelligence.” Sometimes it is just less wasted work. The Engram paper makes an almost irritatingly sensible point. Transformers do two jobs at once: they remember stable patterns, and … Read more
Watch or Listen on YouTube Epiplexity vs. Entropy: The New Metric That Explains What AI Actually Learns Introduction If you’ve ever stared at a loss curve and thought, “Cool, it’s going down, but what did the model actually learn?”, you’re not alone. We’ve trained ourselves to treat lower loss like truth. But a model can … Read more
Watch or Listen on YouTube NVIDIA Rubin Explained: Six Chips, One Rack-Scale AI Supercomputer Introduction Every GPU generation comes with a familiar temptation, stare at the peak numbers, then mentally multiply by “how many can I buy.” It is a comforting game because it avoids the messy part, the part where real systems lose time, … Read more
Watch or Listen on YouTube Recursive Language Models: Introduction A funny thing happens when you give a modern LLM a huge context window. You expect it to feel like upgrading from a studio apartment to a warehouse. In practice, it often feels like moving into that warehouse and then forgetting what you put where. As … Read more
Watch or Listen on YouTube Deep Delta Learning Explained: Introduction ResNet’s big trick is almost insultingly simple. Add an identity shortcut, stop gradients from dying, train deeper models. The shortcut is a safety rail, and in deep residual learning for image recognition it turned “very deep” from a heroic gamble into a default option. That … Read more
Watch or Listen on YouTube The Great Swap: Trust in AI Science Introduction A funny thing happened after ChatGPT shipped. The average paper got easier to read. Not “more correct,” not “more insightful.” Just smoother. The kind of prose that used to signal a careful researcher suddenly became the default setting. That’s why the “AI … Read more
Watch or Listen on YouTube DeepSeek mHC Explained: Stable Hyper-Connections For Wider Transformers Introduction If you have ever watched a large run go sideways at 3 a.m., you know the weird part is not the failure. The weird part is how sudden it is. Loss looks fine, gradients look normal, then a small architectural “improvement” … Read more