reasoning-capability-separable-at-both-training-and-inference
IN derived (depth 2)
Created 2026-06-21T10:12:39+00:00 · Reviewed 2026-06-21T14:41:08+00:00
Explicit reasoning is a separable capability dimension addressable independently at both training time (o1 scoring 83% vs GPT-4o's 13% on math, DeepSeek R1 matching proprietary models via pure RL) and inference time (CoT → self-consistency → tree-of-thoughts) — suggesting reasoning is not simply emergent from scale but a distinct axis that can be optimized orthogonally to model size.
Summary
Reasoning is its own dial you can turn, separate from raw model size. That means a system can get meaningfully better at step-by-step problem-solving either by changing how the model is trained (rewarding logical steps) or by changing how it's prompted at query time (trying multiple paths, backtracking), without needing to build a bigger model.
Justifications
SL — Training-time reasoning models (depth-1) and inference-time structured prompting (depth-1) independently demonstrate reasoning as a separable dimension
Antecedents (all must be IN):
- IN reasoning-models-represent-distinct-capability-tier — Reasoning-specialized models — OpenAI o1 scoring 83% vs GPT-4o's 13% on IMO qualifying problems, DeepSeek R1 matching proprietary models at lower cost — represent a distinct capability tier above standard LLMs, achievable through both proprietary and open-weight approaches.
- IN structured-reasoning-prompting-evolved-from-linear-to-branching — Prompting for reasoning evolved from linear chain-of-thought (single path) to self-consistency (multiple paths, majority vote) to tree-of-thoughts (branching with backtracking), progressively adding search structure.
Dependents
These beliefs depend on this one:
- IN reasoning-and-alignment-are-orthogonal-capability-dimensions — Reasoning capability and alignment quality appear to be separable dimensions in LLM development — reasoning can be independently addressed at both training time (o1, R1) and inference time (CoT, ToT), while alignment has diversified into three independent paradigms (RLHF, DPO/KTO, Constitutional AI) — suggesting that the capability-alignment design space may be at least a 2×2 matrix (training/inference × capability/alignment), with each quadrant potentially offering independent optimization levers.
- OUT structured-reasoning-overcomes-factual-accuracy-gap — The evolution of structured reasoning prompting (CoT → self-consistency → ToT) combined with training-time reasoning specialization provides systematic methods to close the factual accuracy gap between LLMs and humans.