SPURIC SPURICSPUR Innovation Centre
Create free account Start the course
Free • Self-Paced • Certificate on Completion

Master AI - the frontier, explained.

Ten modules that take you inside the models. Transformer internals, how pretraining and post-training actually work, the papers moving the frontier, mixture-of-experts, efficient attention, inference tricks, interpretability, and the safety and governance questions that every serious practitioner has to have an answer for.

Start the course See curriculum All courses
Signup is free. No credit card. Your progress is saved to your account.
Researchers discussing frontier AI at SPUR
Full curriculumMulti-module track
QuizzesEvery module
DiagramsInline mermaid visuals
CertificateOn 100% completion

Who this course is for

You've finished Advanced AI or you've already been running production AI for a while, and now the questions are getting harder. Why does this model do that. What's actually in the paper everyone is citing. What tradeoffs should my team care about when GPT-next ships. This course is for the practitioners who want to reason from first principles, not from Twitter.

Prerequisites

  • Required: Advanced AI (or equivalent: you've shipped AI in production and understand evals, agents, and fine-tuning).
  • Comfort reading technical papers. We'll translate, but the raw material shows up.
  • Basic linear algebra and probability. Refreshers provided where needed.

What you'll learn

Ten modules, three lessons each. Quizzes everywhere. Every mechanism gets a diagram you can actually read.

Module 1 - Transformers

Attention, residual streams, tokenisation, and what's really happening on each forward pass.

Module 2 - Post-training

SFT, RLHF, DPO, and the newer alignment techniques. Why the base model is never the whole story.

Module 3 - Research frontier

A guided tour of the work that's shaping the next generation of models, with the papers that matter.

Module 4 - Pretraining

Data, scale, scaling laws, and what a training run actually costs once you write it all down.

Module 5 - Mixture-of-experts

How MoE works, why it's everywhere now, and the operational gotchas nobody mentions up front.

Module 6 - Efficient attention

FlashAttention, linear attention variants, sparse patterns. The tricks that make long contexts affordable.

Module 7 - Inference optimization

KV caching, speculative decoding, quantisation, and the serving stack tradeoffs that keep cost per token sane.

Module 8 - Interpretability

Mechanistic interpretability, probes, feature attribution. What we actually know about what's inside.

Module 9 - Agentic frontier

The state of the art in autonomous agents, long-horizon tasks, and the capability jumps on the near horizon.

Module 10 - Safety and governance

Catastrophic risks, evaluation frameworks, national policy, and the governance questions every serious team eventually has to answer.

Format

  • Cost: Free. No paywall.
  • Pace: Self-paced - start anytime.
  • Time estimate: Most learners finish in 8-12 weeks at ~4-6 hours/week.
  • Assessments: Quizzes in every module.
  • Diagrams: Inline mermaid diagrams for every architectural concept.
  • Progress: Saved to your account, syncs across devices.

Certificate

Complete all ten modules at 100% and SPUR issues a signed, verifiable PDF certificate in your name. Free. Listed on your SPUR profile and verifiable on spuric.com.

Step to the frontier

Master AI is free. Signup takes 30 seconds. Start Module 1 immediately.

Start the course See all courses