How do test-time resources and training improve model reasoning?

45 specific questions

• Why might latent reasoning capture types of thinking that verbalized CoT cannot?
• Can latent reasoning scale test-time compute without verbalized tokens or special training?
• Can latent reasoning achieve the same substitution without tokens?
• Can continuous latent reasoning match discrete chain-of-thought without training modifications?
• Can latent reasoning scale test-time compute without verbal tokens?
• Can reasoning happen in latent space without chain of thought?
• Can latent reasoning in continuous space scale beyond supervised reasoning tasks?

21 specific questions

• Why does training data format shape reasoning strategy more than domain content?
• How much does training data format influence reasoning strategy versus domain content?
• How does training format shape reasoning strategy more than content?
• Does training data format shape reasoning strategy more than domain content?
• Why does training data format shape reasoning strategy more than content?
• Does training data format shape model reasoning more than domain content?
• Can training format itself shape what reasoning strategy a model learns?

64 specific questions

• Do base models contain latent reasoning that minimal training can unlock?
• Can minimal training signals unlock latent reasoning capability in base models?
• What latent reasoning capability do base models already possess before training?
• Can models possess latent reasoning capability that training signals fail to unlock?
• Can minimal training signals unlock reasoning already latent in pretrained representations?
• Does the base model already contain latent reasoning capability?
• What mechanisms activate latent reasoning capabilities already present in base models?

64 specific questions

• Does scaling reasoning capability create tradeoffs with instruction following?
• Why do models learn reasoning form instead of actual abstract inference?
• Why do instruction following and reasoning capability trade off in training?
• Does reasoning structure match explicit versus implicit task demands?
• Can correct outputs mask reliance on surface heuristics rather than deep understanding?
• Why does instruction-following capability decrease as models scale stronger?
• Can models learn when to think versus answer directly?

27 specific questions

• Can reasoning learned from language modeling actually transfer to knowledge-intensive domains?
• Does model scaling improve knowledge storage faster than reasoning ability?
• Why does reasoning training improve math but hurt knowledge tasks?
• Can mathematical reasoning improvements transfer across problem subdomains?
• What makes knowledge-rich specialized domains structurally different from general reasoning tasks?
• Does task diversity in pretraining data transfer reasoning better than larger models?
• Why does general reasoning not transfer to knowledge-intensive medical domains?

59 specific questions

• Can reasoning traces prove models are actually reasoning versus mimicking?
• Why do reasoning traces fail to accurately reflect model decision-making?
• Can reasoning traces serve purposes beyond producing the final answer itself?
• Why do corrupted traces maintain performance as well as correct traces?
• How do reasoning traces fail to represent what models actually computed?
• What makes a reasoning trace causally sufficient versus merely stylistically plausible?
• Can corrupted reasoning traces be reliably distinguished from correct ones?

29 specific questions

• What causes policy entropy collapse in reasoning-focused reinforcement learning?
• Does policy entropy collapse limit how many iterations of reasoning training work?
• What happens to model reasoning when policy entropy collapses during RL?
• Why does policy entropy collapse limit reasoning and dialogue RL scaling?
• Does policy entropy collapse prevent inference-time search from finding solutions?
• Why does policy entropy collapse when scaling RL for reasoning?
• How does policy entropy collapse constrain token-level distribution in reasoning?

36 specific questions

• Why do language models substitute parametric knowledge over retrieved context mid-reasoning?
• Can models internalize retrieved context as static parametric knowledge?
• Can models recover knowledge with completely unrelated retraining tasks?
• Does finetuning facts into weights overwrite existing model capabilities?
• How does parametric knowledge sabotage context-grounded question answering?
• What makes some contexts learnable as rules versus requiring model retraining?
• How do we distinguish knowledge encoding from knowledge usage in models?

78 specific questions

• Why do reasoning models wander instead of searching systematically?
• What mechanisms cause reasoning models to wander rather than focus?
• How does instance novelty rather than chain length explain reasoning failure?
• Do reasoning models switch approaches when encountering local difficulty?
• Why do reasoning models fail at learning hidden rules from sparse exceptions?
• When does explicit reasoning actually degrade performance on a task?
• Why do reasoning models fail on structurally unfamiliar instances?

38 specific questions

• Why does fine-tuning degrade reasoning quality even as accuracy improves?
• Does supervised fine-tuning improve accuracy while damaging the quality of reasoning?
• Does SFT degrade reasoning quality while improving domain accuracy?
• How does data quality mismatch create reasoning degradation in supervised fine-tuning?
• Can fine-tuning ever teach semantic inference instead of amplifying training shortcuts?
• Why does supervised fine-tuning degrade reasoning quality despite raising accuracy?
• Does supervised fine-tuning improve reasoning or just response formatting?

66 specific questions

• Does structured decomposition improve LLM reasoning in other compound tasks?
• Can language models reason without relying on learned semantic patterns?
• Does more thinking always help large language models or sometimes hurt?
• Why do language models imitate reasoning form without abstract inference capability?
• Can language models reason without relying on surface level pattern matching?
• Do language models build world models or just task-specific heuristics?
• Why does augmenting symbolic reasoning outperform replacing it entirely?

47 specific questions

• What evidence shows that reasoning chains encode token-level functional structure?
• What makes multi-paradigm chaining a distinct reasoning topology?
• Can small models solve complex tasks using externalized reasoning graphs?
• Can single-hop knowledge automatically compose into multi-hop capability?
• Can recursive sub-calls decompose reasoning across multiple context chunks?
• How do humans and LMs differ on multi-hop reasoning?
• Can long-context models handle compositional reasoning requiring structured logic?

50 specific questions

• Do models deliberately hide influences from their reasoning traces?
• Are reasoning models more vulnerable to persuasion than standard models?
• Can activation probes detect reasoning that models omit from text?
• Are reasoning models more vulnerable to adversarial manipulation than standard models?
• Why does latent reasoning override no-think instructions in models?
• How do sycophancy hints stay invisible despite appearing in reasoning chains?
• Do reasoning models fail to report processes that actually influence their answers?