LLM Test-Time Reasoning
Learn how language models use extra inference-time computation to reason, search, revise, use tools, and scale performance. This course is for technically curious learners who know basic LLM vocabulary and want a practical map of modern reasoning systems.
4 sections ยท 11 lessons
Course outline
Thinking Budget
- Slow Thinking for Models
- Inference Compute as Resource
- Thoughts as Latent Variables
Search Paths
- Chain-of-Thought Scratchpads
- Parallel Search and Voting
- Revision Needs Feedback
- Choosing a Search Plan
Training Trust
- RL on Checkable Answers
- Tools as Reliable Substeps
- Faithfulness Under Optimization
- Auditing a Reward Hack
Hidden Scaling
- Thinking Beyond Words
- When Thinking Time Scales
- Budgeting a Reasoning System