Discover AI Courses

Understanding Grokking - A New Performance Phase in Large Language Models

Discover how extended training of LLMs leads to "grokking" - a breakthrough phase where models develop superior generalization abilities and form geometric patterns in embedding spaces.

• YouTube
• 30 minutes
• Self-Paced
• Free Video

Agentic AI: Understanding Advanced Function Calling and RAG Systems

Delve into the advanced concepts of agentic AI systems, exploring function calling, RAG integration, and future world models for enhanced AI planning and decision-making capabilities.

• YouTube
• 46 minutes
• Self-Paced
• Free Video

LongRoPE and Theta Scaling for Extended Context Length in LLMs - Part 2

Delve into advanced techniques for extending LLM context lengths through LongRoPE and Theta Scaling, exploring methods to enhance model performance from 8K to 4M tokens.

• YouTube
• 59 minutes
• Self-Paced
• Free Video

RoPE: Rotary Position Embedding for Extended Context Lengths in Transformers

Dive into Rotary Position Embedding (RoPE) and learn how this innovative technique enhances transformer models by enabling extended context lengths up to 100K tokens.

• YouTube
• 40 minutes
• Self-Paced
• Free Video

In-Context Learning - Comparing Extreme Context Length with Fine-Tuning and RAG

Explore groundbreaking research on in-context learning with extended context models, comparing performance against fine-tuning and RAG methods while examining scaling impacts and demonstration effectiveness.

• YouTube
• 22 minutes
• Self-Paced
• Free Video

Fine-Tuning Large Language Models: Impact on Knowledge Integration and Hallucinations

Delve into groundbreaking research on how fine-tuning Large Language Models with new knowledge affects their tendency to generate hallucinations, exploring methodologies and practical implications.

• YouTube
• 27 minutes
• Self-Paced
• Free Video

Matrix Long Short-Term Memory (mLSTM) - A New Alternative to Transformer LLMs

Explore the innovative xLSTM architecture and its matrix-based approach to neural networks, comparing its potential advantages over traditional transformer models in language processing.

• YouTube
• 23 minutes
• Self-Paced
• Free Video

Understanding DSPy - Programming AI Pipelines for Language Models

Explore DSPy's innovative approach to AI pipeline programming, from its ColBERT v2 origins to practical applications in modular pipeline generation and self-improving language programs.

• YouTube
• 28 minutes
• Self-Paced
• Free Video

Understanding Retrieval Heads in Large Language Models - From Discovery to Applications

Discover groundbreaking research on retrieval heads in transformer architecture, exploring their impact on RAG systems, long-context processing, and chain-of-thought reasoning for enhanced AI capabilities.

• YouTube
• 31 minutes
• Self-Paced
• Free Video

Understanding Snowflake Arctic 480B - A Mixture of Experts LLM Architecture

Dive into the architecture and performance analysis of Snowflake's Arctic 480B LLM, exploring how its 128x4B MoE design compares to traditional transformers for enterprise-level AI tasks.

• YouTube
• 23 minutes
• Self-Paced
• Free Video

Unsupervised In-Context Learning Plus: From Few-Shot to Many-Shot Learning in Large Language Models

Explore advanced In-Context Learning techniques for Large Language Models, from few-shot to many-shot approaches, focusing on unsupervised learning and autonomous capabilities with 1M token context.

• YouTube
• 31 minutes
• Self-Paced
• Free Video

RecurrentGemma: Moving Past Transformers with Griffin Architecture for Long Context Length

Explore Google's RecurrentGemma-2B architecture, featuring Griffin's innovative approach to language modeling that surpasses traditional transformers with enhanced efficiency and long-context processing capabilities.

• YouTube
• 31 minutes
• Self-Paced
• Free Video

TransformerFAM and BSWA: Understanding Feedback Attention Memory and Block Sliding Window Attention

Dive into Google's revolutionary TransformerFAM architecture, exploring how feedback attention mechanisms and working memory enhance AI's ability to process indefinite sequence lengths with improved efficiency and context awareness.

• YouTube
• 25 minutes
• Self-Paced
• Free Video

INFINI Attention: Efficient Infinite Context Transformers with 1 Million Token Context Length

Explore Google's Infini-attention transformer mechanism, featuring compressive memory integration for handling million-token sequences and efficient long-term information processing.

• YouTube
• 21 minutes
• Self-Paced
• Free Video

Ring Attention and Blockwise Transformers for Extended Context Length in Language Models

Dive into the technical mechanics of Ring Attention, exploring how it achieves million-token context lengths in large language models through blockwise parallel transformers and efficient implementation.

• YouTube
• 25 minutes
• Self-Paced
• Free Video