Directory
Curated and current: the tools worth knowing, and material that goes deeper than the book.
Grant Sanderson's animated video series explains how neural networks learn, from a single neuron through backpropagation up to how a GPT-style transformer predicts the next word. No coding required; the best visual intuition for beginners.
DetailsSebastian Raschka's newsletter with technical deep dives into how specific LLM architectures and training techniques work, plus regular research roundups. Most core posts are free.
DetailsA one-hour, general-audience talk explaining what LLMs are, how they are trained and where the technology is heading, using analogies rather than code or math. A natural next step right after an introductory LLM book.
DetailsA video series that builds neural networks, a character-level language model and eventually a small GPT from scratch in Python, line by line. Intermediate to advanced, for readers who want to implement what a beginner book only described.
DetailsA live leaderboard where users compare responses from different AI models head to head and vote, producing rankings based on human preference rather than a fixed test set. A beginner-friendly way to see which models are currently rated best.
DetailsAn independent benchmarking site comparing over 250 LLMs side by side on intelligence, price, speed and context window, updated continuously as new models ship. Useful for putting numbers behind model claims.
DetailsThe 2017 paper that introduced the Transformer architecture underlying essentially every modern LLM. Dense original research, best read after one of the visual explainers, but worth having as the primary source.
DetailsA free, self-paced German-language introduction to AI from the University of Helsinki, requiring no math or programming background and completed by over 2 million people worldwide. Broader than just LLMs, a well-established starting point.
DetailsJeremy Howard's free, code-first deep learning course that has students training real models in the first lesson before circling back to theory. Intermediate level, some Python helpful, eventually covers the building blocks behind language models.
DetailsA free, hands-on, 12-chapter course from Hugging Face that teaches how to use and fine-tune real transformer models with their open-source libraries. Requires basic Python, for readers who want to start building.
DetailsA daily-updated feed of the AI research papers gaining the most attention, each linked to its code, discussion and arXiv page. Effectively the successor to the retired Papers with Code.
DetailsA weekly newsletter by Anthropic co-founder Jack Clark covering AI research papers, policy and where the field is headed. Denser and more research-focused than The Batch, widely read among practitioners.
DetailsA free German-language course (about 10 hours, with an English toggle) explaining the Transformer architecture behind modern LLMs, covering attention conceptually without code or heavy math. One of the few solid LLM explainers written natively in German.
DetailsBrendan Bycroft's zoomable 3D walkthrough of a working GPT-style network, tracing every matrix multiplication from input tokens to output logits. More granular than Transformer Explainer, suited to intermediate readers.
DetailsA frequently updated blog tracking how LLMs actually behave in practice, through plain-language write-ups of new model releases, prompt injection risks and open tools. Trusted for being hype-free.
DetailsAndrew Ng's weekly newsletter summarizing that week's AI research and industry news in plain language with a short personal commentary. Beginner-friendly, a good habit for staying current.
DetailsJay Alammar's visual walkthrough of the Transformer architecture behind GPT and other LLMs, using diagrams instead of equations to show attention and encoder-decoder layers. The natural first stop after finishing a beginner LLM book.
DetailsAn interactive, in-browser tool from Georgia Tech that runs a live GPT-2 model on text you type, visualizing tokenization, embeddings, attention heads and next-token prediction as they happen. Built for people with no programming background.
Details