Module Check: How Frontier AI Is Built
You have covered the full arc of how frontier AI is built — from the anatomy of the labs that build it, through the compute, data, and training pipelines, through the economics and evaluation that shape deployment decisions, through the competitive race dynamics and the contested debate over open versus closed release. This lesson consolidates the module. Work through the flashcards to lock in vocabulary, then demonstrate your command of the ideas through the quizzes and final synthesis.
Flashcards — click each card to reveal the answer
Module Quiz
A frontier lab's data team is assembling a training corpus and must decide how much weight to give Wikipedia relative to raw web crawl data. Wikipedia contains far fewer total tokens. Which principle from this module best justifies giving Wikipedia disproportionately high weight?
A base model and an RLHF-trained model are given the identical prompt: 'Explain the water cycle to a 10-year-old.' The base model outputs a graduate-level textbook excerpt on hydrology. The RLHF model outputs a clear, engaging, age-appropriate explanation. Which concept most precisely explains this difference?
A frontier lab's systems engineering team discovers their distributed training run has a GPU utilization of only 58% — far below the theoretical maximum. Which explanation is most technically plausible based on what you learned in this module?
A frontier lab is deciding whether to release the weights of a new model that performs at the level of the previous state-of-the-art. Safety researchers argue against open release; engineering advocates for it. Which argument from the restricted-release side is logically strongest?
A small nation's government wants to develop domestic AI capability at the frontier level and announces a plan to build a cluster of 20,000 advanced AI accelerator GPUs. Which structural constraint from this module most directly threatens this plan?
A venture investor is evaluating two frontier labs. Lab A has achieved state-of-the-art results but has signed an enterprise contract with a cloud provider that gives the provider preferential pricing, first integration rights, and a board observer seat. Lab B has raised only venture capital with no infrastructure partnerships. Which concern is most specific to Lab A's structure?
Synthesis Challenge
Write the Briefing Memo
- You have been hired as a senior policy analyst at a national AI advisory body. Your director has asked for a two-page briefing memo on the following question: 'What are the three most important structural features of how frontier AI is built, and what policy interventions — if any — are justified by each?'
- Your memo should:
- Paragraph 1 — Introduction (50 words): State the purpose of the memo and your three chosen structural features. Be specific — do not choose vague features like 'it is complex.' Draw directly from module concepts.
- Paragraph 2 — Feature 1 (100 words): Describe the first structural feature precisely. Explain why it matters for policy. Identify one specific policy intervention that addresses it and one risk that intervention carries.
- Paragraph 3 — Feature 2 (100 words): Same structure as Paragraph 2 for your second feature.
- Paragraph 4 — Feature 3 (100 words): Same structure as Paragraph 2 for your third feature.
- Paragraph 5 — Conclusion (50 words): Summarize your three interventions and identify the single most important one, with a one-sentence justification.
- Strong memos will: use precise terminology from the module, avoid vague generalities, acknowledge the costs and risks of proposed interventions, and demonstrate that you understand the structural reasons why interventions may or may not be effective. This is the kind of analytical writing that matters in AI policy, technology journalism, research, and industry strategy.