[cosmicbboy]

Reorganizes the tutorials so that Reinforcement learning holds every RL tutorial and Frontier AI covers the workloads frontier AI labs run day to day. Adds five Frontier AI tutorials, each validated end to end on a Union cluster.

Stacked on #1938 (the reinforcement learning tutorials): merge that first.

SkyRL moves into Reinforcement learning

• tutorials/frontier-ai/agentic-rl-skyrl is now tutorials/reinforcement-learning/agentic-rl-skyrl. The move is a rename-only commit, and the user-guide links are updated.
• The RL section is now visible in both variants, because SkyRL has an open-source version. The Union-only tutorials are gated in the section's table and cards.
• The redirects are in Redirects for the SkyRL tutorial move into Reinforcement learning unionai-docs-infra#363, and this PR moves the infra pointer to it.

New Frontier AI tutorials. The section index now opens with the three kinds of work these labs do (data factories, decisions from many small runs, continuous measurement) and a table of who does each workload.

Page	Use case	Result from the validated run
Refine a pretraining corpus	Filters, dedup and decontamination on a CPU fan-out; LLM quality labels distilled into a cheap classifier	300k FineWeb documents to 3,454. The classifier is 66x cheaper per document than the LLM. A new threshold reruns in 33 s from cache
Evaluate every checkpoint	Checkpoints published as artifacts mid-training; an OnArtifact trigger runs a cached benchmark suite on each one	GSM8K 3% → 24%. A guard picks the release candidate. A backfill reruns only the new benchmarks
Curate a video dataset for generative video	Scene cuts, GPU scoring and VLM captions on separate pools; the kept clips published as a dataset artifact	569 candidate clips from CC-BY Blender films, 369 kept
Find features in a language model with sparse autoencoders	One cached activation harvest, 18 SAEs trained in parallel, and Claude labels scored for detection	93–98% loss recovered. Detection score 0.73 against 0.51 for mismatched labels
Closed-loop molecular discovery	A surrogate picks molecules, an xTB oracle measures them on a CPU pool, and a human approval gate sits before each batch	31 hits against 4 for random search on the same oracle budget

Process

• I chose the five use cases from research into what frontier AI labs publish and hire for, then cut and critiqued the candidates for diversity and feasibility.
• After the first validated runs, an independent review found code/doc mismatches and cache-key gaps. All of them are fixed, and the affected tutorials were rerun.

Pointers. The unionai-examples pointer moves to the commit in unionai/unionai-examples#349, and unionai-docs-infra to the commit in unionai/unionai-docs-infra#363. Both should move to their repos' main once those PRs merge.

Checks. markdownlint, cspell, make check-links, check-images, check-subpage-cards, check-icon-names and check-deleted-pages pass, and make dist builds both variants.

🤖 Generated with Claude Code

[github-actions]

GHA build & deploy preview

Built by .github/workflows/build-pr.yml and deployed to the docs CF Pages project by .github/workflows/deploy-pr-preview.yml.

Branch alias	https://pr-1949-niels-frontier-ai-tu.docs-dog.pages.dev
This commit	https://9a7ce151.docs-dog.pages.dev
Commit SHA	d550718e3b4adca54afcdfabbeda0f8847edace4

Updated automatically on every push.