AI ENGVisual Encyclopedia

MODULE 5: REASONING & VERIFIABLE REWARDS · SCENE 15

Chain-of-Thought & Test-Time Compute

System 1 vs. System 2, self-consistency majority voting, and pass@k scaling on MATH and HumanEval.

SAMPLES kSELF-CONSISTENCY · 40%pass@1 · 40%CoT COST · ~1,500 TOKENS
ACCURACY vs TEST-TIME COMPUTE (SINGLE-SHOT 40%)

One chain of thought: the baseline. RUN the slider right to buy accuracy with votes.

TECHNICAL BREAKDOWNModule 5: Reasoning & Verifiable Rewards

Chain-of-thought and test-time compute

Reasoning models spend inference-time tokens to buy accuracy: generate multiple chains of thought, vote or verify, and the effective answer quality climbs with compute — a curve that did not exist for one-shot completions.

System 1 vs System 2

Direct answers are fast intuition; long CoT is deliberation. RLVR training teaches the model when to deliberate and how to backtrack within a single generation.

Self-Consistency

Sample k chains, take the majority answer. Accuracy rises with k with diminishing returns — majority voting cancels independent reasoning errors.

Pass@k

The probability that at least one of k samples is correct. RLVR training improves both the single-sample accuracy and the tail — pass@k curves shift up wholesale.

MATHEMATICAL FORMULATION · TEST-TIME SCALING
pass@k = 1 − C(n−k, c) / C(n, c); self-consistency ≈ p^(1/(k·0.55))

Unbiased pass@k over n graded samples with c correct. If single-sample accuracy is 40%, majority voting over 8 samples behaves like a much sharper estimator.

REAL-WORLD PRODUCTION ENGINEERING
  • Reasoning tokens cost real latency and money: a 4k-token CoT at 50 tok/s adds 80 seconds per answer. Production routers send easy queries to a short-CoT model.
  • Length-controlled evals matter: naive RLVR can inflate CoT length to mine format rewards without improving correctness.