AI Courses
Curated university courses on LLMs, Transformers, Graph Models, Reasoning, and AI Safety — sorted by recency and quality.
Tier 1: Premier Research-Oriented Courses (2025–2026) Premier
Stanford CS336: Language Modeling from Scratch (Spring 2025)
Instructors: Percy Liang & Tatsunori Hashimoto Course site · GitHub
A deep, implementation-heavy graduate course on building LLMs entirely from the ground up. Covers data collection/cleaning, tokenizer and Transformer architecture implementation, model training at scale, and evaluation/deployment. Advanced topics: custom FlashAttention kernels, multi-GPU distribution, scaling laws, and alignment (SFT + RLHF).
| Aspect | Rating | Notes |
|---|---|---|
| Depth | Low-level LLM internals; few courses match this | |
| Relevance | State-of-the-art training techniques | |
| Accessibility | Materials public; requires advanced skills + GPUs | |
| Career Value | Ideal for AI researchers and expert engineers |
Harvard CS 2881: AI Safety (Fall 2025)
Instructor: Boaz Barak Course site · YouTube playlist
Graduate-level course on AI alignment and safety. Covers generalization, optimization, robustness, interpretability, and alignment. Seminar-style with readings, midterm, and NeurIPS-style research final project. Student projects produce original research.
| Aspect | Rating | Notes |
|---|---|---|
| Depth | Conceptually deep, tackles unsolved problems | |
| Relevance | Critical for future AI development | |
| Accessibility | Public lectures and materials | |
| Career Value | For researchers pursuing AI safety |
CMU 11-667: Large Language Models: Methods and Applications (Fall 2025)
Instructors: Jaromir Savelka & Yubin Kim Course site
Holistic graduate survey of LLM techniques. Part 1: architectures, pre-training, inference, evaluation, alignment, emergent capabilities. Part 2: scaling, efficiency, deployment (latency, cost, safety). Students implement and fine-tune models from scratch in PyTorch.
| Aspect | Rating | Notes |
|---|---|---|
| Depth | Thorough on core concepts and applications | |
| Relevance | Covers everything needed for LLM work in 2025 | |
| Accessibility | Syllabus public; lectures not openly shared | |
| Career Value | Excellent for NLP/LLM-focused roles |
CMU 11-868: Large Language Model Systems (Spring 2025)
Instructor: Lei Li Course site
Advanced systems engineering for LLMs. Covers efficient training algorithms, distributed training, memory optimization, inference serving, caching, model sharding, RLHF at scale, and continuous learning. Intersection of ML and systems research.
| Aspect | Rating | Notes |
|---|---|---|
| Depth | Deep dive into LLM engineering | |
| Relevance | Critical as LLMs move to production | |
| Accessibility | Specialized content not openly provided | |
| Career Value | Gold for AI infrastructure engineers |
Stanford CS372: AI for Reasoning, Planning, and Decision Making (Spring 2025)
Instructor: Edward Y. Chang Course site · Stanford Online
Cutting-edge course on reasoning frameworks atop LLMs, targeting AGI-like abilities. Topics: multi-agent LLM systems, planning algorithms, temporal reasoning, validated decision-making, long-term memory, ethical alignment. Research-focused seminar on frontier AI.
| Aspect | Rating | Notes |
|---|---|---|
| Depth | Forward-looking, tackles unsolved problems | |
| Relevance | Highly relevant for future AI, less for current | |
| Accessibility | Limited; some lectures shared informally | |
| Career Value | Goldmine for next-gen AI systems |
Stanford CS224W: Machine Learning with Graphs (Fall 2025)
Instructor: Jure Leskovec Course site · Videos (2021)
Comprehensive deep dive into graph ML and GNNs. Covers PageRank, community detection, representation learning, GNN architectures, Graph Transformers, knowledge graph reasoning. Applications: social networks, recommender systems, biology.
| Aspect | Rating | Notes |
|---|---|---|
| Depth | One of the most in-depth graph ML courses | |
| Relevance | Crucial for relational data domains | |
| Accessibility | Materials free; older recordings available | |
| Career Value | Foundational for graph ML research |
CMU Advanced NLP (Fall 2025)
Course site · YouTube playlist
Graduate-level, practice-oriented coverage of modern NLP: transformers and LLMs, prompting and agents, alignment (RL/GRPO), evaluation, safety, and applied projects.
Tier 2: Excellent Foundational Courses (2025) Excellent
Stanford CME 295: Transformers & Large Language Models (Autumn 2025)
Instructors: Afshine Amidi & Shervine Amidi Course site · Cheatsheet · Syllabus + Videos
Comprehensive intro to Transformers and LLMs. Covers evolution of NLP, Transformer architecture (tokenization, embeddings, attention), LLM foundations (MoEs, decoding), training/tuning (SFT, RLHF, LoRA), RAG, and agentic workflows.
| Aspect | Rating | Notes |
|---|---|---|
| Depth | Great breadth, solid depth for overview | |
| Relevance | Ubiquitous transformer paradigm | |
| Accessibility | Free lectures, slides, illustrated cheatsheet | |
| Career Value | Strong for practitioners and NLP foundations |
Stanford CS 120: Introduction to AI Safety
Instructor: Max Lamparth Course site
Foundational AI safety course covering interpretability, robustness, RL safety, NLP issues. Topics span reinforcement learning, computer vision, and NLP. No strict prerequisites.
| Aspect | Rating | Notes |
|---|---|---|
| Depth | Solid foundations | |
| Relevance | Critical safety topics | |
| Accessibility | Open to Stanford community | |
| Career Value | Good entry point for AI safety |
Stanford CS329H: Machine Learning from Human Preferences
Course site · YouTube playlist
Advanced topics: preference modeling, RLHF/GRPO, reward modeling, dataset construction, safety and alignment, evaluation beyond accuracy, applications to LLMs and agents.
MIT MAS.S60: How to AI (Almost) Anything (Spring 2025)
Instructor: Paul Liang (MIT Media Lab) Course site · Videos
Broad modern AI survey emphasizing foundation models and multimodal applications. Covers LLMs plus vision, audio, sensor data, music, art. Topics: multimodal fusion, cross-modal mapping, grounding AI in physical world.
| Aspect | Rating | Notes |
|---|---|---|
| Depth | Wide range at solid graduate level | |
| Relevance | Foundation models + multimodal AI | |
| Accessibility | Free via MIT OCW | |
| Career Value | Panoramic view of modern AI |
Davidson CSC 381: Deep Learning (Fall 2022)
Instructor: Prof. Bryce Wiedenbeck YouTube playlist
"Hands down, the clearest and most effective explanation of Transformers on YouTube." — Highly praised lecture series covering neural networks, deep learning fundamentals, and an exceptionally clear treatment of Transformers and self-attention.
| Aspect | Rating | Notes |
|---|---|---|
| Depth | Excellent pedagogical clarity | |
| Relevance | Timeless Transformer foundations | |
| Accessibility | Free on YouTube | |
| Career Value | Best starting point for understanding Transformers |
Tier 3: Specialized & Domain Courses Specialized
CS-E4740: Federated Learning (Aalto University, Spring 2025)
Instructor: Alexander Jung Course site · GitHub · YouTube playlist
FL networks, design principles, core algorithms, variants, and trustworthy FL. Weekly lectures and Python/Jupyter exercises.
HAII-2024: Human-AI Interaction (Politecnico di Torino)
Instructors: Luigi De Russis, Alberto Monge Roffarello YouTube playlist
Ph.D. course on methods for designing, building, and evaluating human–AI interactions. Covers prototyping, UX for AI, evaluation methods, and case studies.
HAI Fall Conference 2022: AI in the Loop (Stanford HAI)
Human-in-the-loop AI decision-making processes. Explores keeping humans at the center of AI technologies.
Professional Programs & Certificates Professional
CMU Graduate Certificate: Generative AI & Large Language Models (Online, 2025)
12-month online program (3 courses). Build LLMs from scratch, fine-tune models, distributed training (SLURM), RLHF, compression, multimodal AI. Live evening classes for working professionals.
| Aspect | Rating | Notes |
|---|---|---|
| Depth | Compressed master's-level curriculum | |
| Relevance | Tuned to 2025 industry needs | |
| Accessibility | Online but paid (~tuition-based) | |
| Career Value | CMU credential + immediately applicable skills |
MIT Professional: AI System Architecture & LLM Applications (July 2026)
Instructor: David Martinez (MIT Lincoln Laboratory) Program site
5-day intensive on designing, building, and deploying LLM-powered systems end-to-end. Covers architecture design, deployment at scale, testing, evaluation, responsible AI. Culminates in group project building an LLM application.
| Aspect | Rating | Notes |
|---|---|---|
| Depth | Thorough overview for practitioners | |
| Relevance | LLM deployment — critical skill | |
| Accessibility | Paid (~$4,500), in-person Cambridge | |
| Career Value | Fast-track for technical leads/managers |
Oxford (LMH): Advanced AI & ML: LLMs and NLP (Summer 2026)
3-week intensive (in-person or online). Covers Transformers, attention, LLM fine-tuning (LoRA, QLoRA), multimodal AI (vision-language), AI ethics. Hands-on labs.
| Aspect | Rating | Notes |
|---|---|---|
| Depth | Rigorous for summer program | |
| Relevance | LLMs, NLP, vision-language | |
| Accessibility | Online option; paid program | |
| Career Value | Oxford credential + networking |
Quick Reference by Topic
| Topic | Top Recommendations |
|---|---|
| Transformers (from scratch) | Davidson CSC 381, Stanford CME 295 |
| LLM Training & Systems | Stanford CS336, CMU 11-868 |
| LLM Applications | CMU 11-667, MIT MAS.S60 |
| AI Safety & Alignment | Harvard CS 2881, Stanford CS 120, Stanford CS329H |
| Reasoning & AGI | Stanford CS372 |
| Graph ML & GNNs | Stanford CS224W |
| Multimodal AI | MIT MAS.S60, Oxford Summer |
| Professional Upskilling | CMU Certificate, MIT Professional |
Learning Path Recommendations
For beginners wanting to understand Transformers: 1. Davidson CSC 381 (clearest explanations) 2. Stanford CME 295 (comprehensive + cheatsheet)
For researchers building LLMs: 1. Stanford CS336 (build from scratch) 2. CMU 11-868 (systems engineering) 3. CMU 11-667 (methods + applications)
For AI safety focus: 1. Stanford CS 120 (foundations) 2. Harvard CS 2881 (research-level) 3. Stanford CS329H (human preferences)
For industry practitioners: 1. Stanford CME 295 (fast ramp-up) 2. CMU Certificate (structured + credential) 3. MIT Professional (intensive deployment)
Last updated: January 2026