Test-Time Compute

Definition

Test-time compute (TTC) refers to the computational resources spent during inference to improve model outputs, as opposed to resources spent during training. The key insight: models can “think longer” on hard problems rather than always taking the same number of compute steps per token.

Why It Helps

Three complementary explanations from Lilian Weng:

  1. Compute as resource: CoT dramatically increases FLOPs per answer token, allowing variable compute proportional to problem difficulty.
  2. System 1 → System 2: Mirrors Kahneman’s dual-process theory — fast autoregressive prediction = System 1; deliberate reasoning steps = System 2.
  3. Latent variable modeling: Thoughts z are hidden variables that enrich P(y|x) — sampling multiple chains of thought is sampling from the posterior P(z|x,y).

Main Approaches

ApproachMechanismLimitation
Best-of-N (parallel)Generate N outputs, pick bestBounded by model capability
Self-consistencyMajority vote across CoT chainsNo external verification
Sequential revisionIteratively self-correctRequires external feedback; naive self-correction often fails
Process reward modelsStep-by-step scoringExpensive to train
RL on checkable tasksPolicy gradient on verifiable rewardsOnly works for problems with ground truth

The RL Path to Better Reasoning

The biggest TTC gains came from RL training on problems with automatically verifiable answers (STEM, coding with unit tests). This produced:

  • OpenAI o1, o3
  • DeepSeek-R1 (policy gradient only, no supervision needed)

See RL Infrastructure for the training side.

Open Questions

  • Faithfulness: Do CoT chains reflect actual computation, or are they post-hoc rationalization? Evidence suggests models often explain reasoning differently from how they compute. See How Claude Thinks.
  • Scaling laws: How does test-time compute scale with problem difficulty? Larger models benefit more from thinking time.
  • Continuous space thinking: Is discrete token CoT the right substrate? Alternatives: recurrent architectures, latent variable approaches.