- formatting
- images
- links
- math
- code
- blockquotes
- external-services
•
•
•
•
•
•
-
Sparse Autoencoders: The Dictionary of Concepts Inside LLMs
How sparse autoencoders are helping researchers discover millions of monosemantic features inside large language models — a breakthrough in AI interpretability.
-
Multimodal Foundation Models: Teaching AI to See and Read Together
CLIP, LLaVA, Flamingo, and GPT-4V — how modern AI systems fuse vision and language into unified world representations.
-
Neural Scaling Laws: The Power Laws Governing Every LLM
Kaplan's and Chinchilla's scaling laws demystified — the power laws every major LLM training run is designed around.
-
Chain-of-Thought: Why Thinking Out Loud Makes AI Smarter
Chain-of-thought prompting, self-consistency, Tree-of-Thoughts, and the new era of reasoning models that scale test-time compute.
-
Retrieval-Augmented Generation: Grounding LLMs in Facts
How RAG systems combine dense vector retrieval with language model generation to produce factually grounded, up-to-date answers.