Skip to content
InfratoApp

Anthropic · Course

AI capabilities and limitations

AI capabilities and limitations gives learners a mental model of how generative AI behaves, organised around four properties: next token prediction, knowledge, the context window and steerability. It is the companion to AI Fluency: Framework and foundations.

Open at Anthropic ↗Common questions

Who it suits

  • Anyone who finds AI outputs delightful one moment and confusing the next, and wants to know when to trust them.
  • Learners who have taken or plan to take the AI Fluency course: Anthropic says the two can be taken in either order.

What it covers

  • Generative AI versus classification and prediction AI, and the four core properties as capability-to-limitation continuums.
  • Pretraining and fine-tuning, and the behaviours they leave: sycophancy, verbosity, over-caution and loose confidence calibration.
  • Next token prediction, and where fabrication concentrates.
  • The knowledge cutoff, the context window as a hard limit, and context-placement strategies.
  • Diagnosing real-world AI failures and connecting the properties to the 4D framework.

Common questions about AI capabilities and limitations

How long is AI capabilities and limitations?
Anthropic lists 13 lessons on the course page.
How does it relate to AI Fluency?
Anthropic says AI Fluency teaches the human competencies, while this course teaches the machine properties they respond to, and they can be taken in either order.
Is it technical?
It explains how models work conceptually, to help you judge outputs and choose fixes, without requiring you to write code.
What is the practical payoff?
Calibrated trust: locating a task on each continuum and picking a targeted fix when something fails.

More like this

Keep learning on Infrato

Where this comes from

Written from Anthropic’s own pages, read on 2026-10-13. Content, prices and availability change, so the provider’s page is the source. We are not affiliated with Anthropic and this is not an endorsement.