Yannic Kilcher Courses

LLaMA- Open and Efficient Foundation Language Models - Paper Explained

1 review

Detailed explanation of Meta's LLaMA language models, covering training data, architecture, implementation, and performance. Explores how smaller models can outperform larger ones with optimized training.

• YouTube
• 41 minutes
• On-Demand
• Free Video

Tree-Ring Watermarks - Fingerprints for Diffusion Images That Are Invisible and Robust

Explores a novel watermarking technique for diffusion models that embeds invisible, robust fingerprints in generated images, discussing its implementation, effects, and potential applications beyond copyright protection.

• YouTube
• 36 minutes
• On-Demand
• Free Video

OpenAI Fires CEO Sam Altman - Leadership Transition and Industry Impact

Análisis detallado de la sorpresiva destitución de Sam Altman como CEO de OpenAI, explorando las posibles razones, reacciones y consecuencias para la empresa y la industria de IA.

• YouTube
• 20 minutes
• On-Demand
• Free Video

Q-Learning: From Basics to Deep Q-Networks

Explore Q-Learning fundamentals, from Markov Decision Processes to Deep Q-Networks, in this comprehensive overview of reinforcement learning concepts and applications.

• YouTube
• 46 minutes
• On-Demand
• Free Video

Scalable Extraction of Training Data from Language Models

Explore techniques to extract training data from language models, revealing vulnerabilities in data privacy and model security. Learn about new attacks and their implications for AI alignment and memorization.

• YouTube
• 48 minutes
• On-Demand
• Free Video

Text Embeddings Reveal Almost As Much As Text - Iterative Embedding Inversion

Explore how text embeddings can potentially reconstruct original text, discussing methods, implications, and prevention strategies for this emerging privacy concern in natural language processing.

• YouTube
• 37 minutes
• On-Demand
• Free Video

Google's Gemini Video Controversy - Analysis and Implications

Analysis of Google's Gemini marketing video, examining its portrayal of AI capabilities and raising questions about authenticity and potential misrepresentation of the model's true abilities.

• YouTube
• 16 minutes
• On-Demand
• Free Video

NeurIPS 2023 Research Highlights - Circuits, Collaborative Filtering, Clustering, and Privacy

Overview of NeurIPS 2023 poster presentations covering graph neural networks, collaborative filtering, semi-supervised learning, and privacy assessment in image reconstruction, with insights on cutting-edge machine learning research.

• YouTube
• 34 minutes
• On-Demand
• Free Video

NeurIPS 2023 Research Highlights - Poster Session 2

Explore cutting-edge AI research from NeurIPS 2023, covering topics like RNN training, image editing, language models, transformers, geo-localization, and hardware resilience in image classifiers.

• YouTube
• 44 minutes
• On-Demand
• Free Video

Mamba: Linear-Time Sequence Modeling with Selective State Spaces

Explore Mamba, a novel linear-time sequence modeling architecture using selective state spaces. Learn about its advantages over Transformers and its applications in language, audio, and genomics.

• YouTube
• 41 minutes
• On-Demand
• Free Video

NeurIPS 2023 Poster Session 4 - Highlights in Machine Learning Research

Explore cutting-edge machine learning research in temporal action segmentation, test-time adaptation, recurrent neural networks, dictionary learning, equivariant adaptation, cross-task generalization, and adversarial learning.

• YouTube
• 58 minutes
• On-Demand
• Free Video

GPT-4 Solves MIT Exam with 100% Accuracy - OpenLLaMA 13B Released

Análisis crítico de un estudio sobre GPT-4 resolviendo exámenes del MIT, revelando problemas metodológicos y discutiendo las implicaciones para la evaluación de modelos de IA en educación.

• YouTube
• 31 minutes
• On-Demand
• Free Video

Reinforced Self-Training for Language Modeling - Paper Explained

Explores Reinforced Self-Training, a novel approach for improving language models using offline reinforcement learning, offering efficiency advantages over traditional online methods.

• YouTube
• 53 minutes
• On-Demand
• Free Video

Building a CPU Using GPT for Every Instruction

Explore building a CPU using GPT for every instruction, compiling C code to LLVM-IR. Learn about compilers, virtual machines, and innovative applications of language models in computing.

• YouTube
• 22 minutes
• On-Demand
• Free Video

Retentive Network - A Successor to Transformer for Large Language Models

Explores Retentive Network as an alternative to Transformer, offering training parallelism and efficient inference for large language models. Discusses retention mechanism, computation paradigms, and experimental results.

• YouTube
• 28 minutes
• On-Demand
• Free Video