Reasoning and Verifiable Learning

Covers reasoning types, reasoning in LLMs, reasoning failure modes, reasoning evaluation.

Covers CoT prompting, zero-shot CoT, CoT fine-tuning, CoT limitations.

Covers self-consistency, tree of thought, least-to-most prompting, decomposition strategies.

Covers step verification, process reward models, verification-guided search, self-correction.

Covers math problem solving, symbolic integration, math benchmarks, math reasoning training.

Covers systematic failures, spurious correlations, reasoning shortcuts, robustness challenges.

Covers o1-style reasoning, test-time compute scaling, reasoning-capable models, open research questions.

Rule-based and executable rewards, correctness verification, reward design, and domains where RLVR works.

GRPO-style objectives, relative advantages, stability, sampling costs, and implementation choices.

Bootstrapping reasoning behavior, filtering traces, iterative training, and transferring reasoning to smaller models.

Sparse outcomes, step-level supervision, verifier reliability, credit assignment, and reward hacking.

Sampling, verifier-guided search, self-correction, budget allocation, stopping, and compute-optimal reasoning.

When visible chains of thought are explanations, when they are not, and alternatives for latent deliberation.

Unnecessary deliberation, error amplification, confidence, adaptive budgets, and concise reasoning.