Learn Frontier AI with any AI

The live areas of work, in rough order of how the field is moving.

• Scaling & reasoning models — larger systems, and models that "think" before answering.
• Agentic AI — systems that don't just answer but plan, use tools and act over many steps.
• The AGI debate — what "general" intelligence would mean, and how close we are.
• Alignment & safety — making powerful systems reliably do what we intend.
• Interpretability — trying to understand what's actually happening inside a model.
• Work & the economy — what capable AI does to jobs and prosperity.
• Governance — regulation, and the geopolitics of compute and capital.

A few fixed points — but for anything recent, the Frontier prompt with search will beat any static list.

• The transformer (2017) — the architecture that made modern large language models possible.
• The scaling era — the discovery that capability climbs fairly predictably with size, data and compute.
• Learning from human feedback — the technique that turned raw text predictors into usable assistants.
• The turn to reasoning and agents — models that deliberate and take actions, not just complete text.
• The safety and alignment field — a serious research community working on controlling advanced systems.
• AI governance efforts — a fast-evolving landscape of national and international rules.

This is the most contested subject in the library. A good tutor keeps both sides at full strength.

• How close is AGI — and would we know it? Estimates range from a few years to never, held by serious people.
• Is alignment tractable? From "a hard but normal engineering problem" to "the defining challenge of the century."
• Existential risk: real or a distraction? A genuine split among informed observers.
• Open or closed frontier models? Openness for scrutiny and access, versus restriction for safety.
• Concentration of power. What it means that frontier AI depends on scarce compute and vast capital.

A route in — and here, order matters more than usual.

1. Turn on web search and run The Frontier first. On this node, static knowledge goes stale fast — start from what's true this month.
2. Use Great Debates on AGI timelines or existential risk to hear both sides properly.
3. Run Orientation on scaling laws or agentic AI for the underlying concepts.
4. Then follow the actual labs and safety institutes directly — this field is a moving target.

Hold the uncertainty honestly: confident predictions here, in either direction, should make you suspicious.