Skip to content
Grok Academy

course

AI Capabilities and Limitations

Build an accurate mental model of what large language models can and cannot do: next-token prediction, knowledge, working memory, steerability, and context limits.

4 lessons · 1 quiz · 58 min

For: Anyone who will rely on Grok for non-toy work. No math required.

Start course

What you’ll walk out with

  • Explain next-token prediction in plain language
  • Separate training knowledge from context you provided
  • See why models hallucinate, flatter, and forget
  • Work with context windows instead of against them

Lessons

  1. 01Next-token predictionThe engine under the chat UI is not a database and not a mind.12 min
  2. 02What Grok 'knows'Weights vs. this conversation vs. the live web.12 min
  3. 03Working memory and the windowLong context is real. Attention is not uniform. Old instructions fade.12 min
  4. 04Steerability and failure modesHallucination, sycophancy, bias, and jailbreak theater.14 min
  5. QCapabilities quiz4 questions