The foundational technical building blocks that determine what AI systems can learn, reason about, remember, and execute.

5 Mins - Shorts

https://www.youtube.com/playlist?list=PLMgr0xgBW16wfPGaejlOXbYjXPcVlF4iz

30 Mins - Relevant Almost Human Episodes

1. Gilad Lotan Takes on Embeddings: The Unsexy AI Layer That Changes Everything

Embeddings as a representation layer and “meaning → math”:

• embeddings as vectors in high-dimensional space
• cosine similarity / vector arithmetic (“king - man + woman = queen”)
• semantic similarity beyond keywords / TF-IDF / tags
• clustering in embedding space + pitfalls of naive averaging
• vector DB / vector store cost structure, update cadence, static vs dynamic embeddings

2. Liran Tam Takes on Meta Learning: How to Beat Big Tech Without Big Data

Meta-learning as a training paradigm:

• “teach AI how to learn” / rapid adaptation
• few-shot / low-data regimes; adjacent distributions
• inner-loop / outer-loop optimization, meta-parameters
• transfer learning in RL / robotics, cross-domain adaptation