Line of inquiry
Inquiring lines›How do language models develop gen…›How do inference-time strategies a…›this line of inquiry
Do base models contain latent reasoning capabilities that training can activate?
A broader line of inquiry — a family of 50 specific questions the research asks around this. Follow one into its inquiring-line page, or move sideways to a related line below.
Questions in this line of inquiry 50
Specific inquiring lines the field asks around this — ordered from the most general framing down to the most specific angle.
- 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?
- What mechanisms activate latent reasoning capabilities already present in base models?
- Can models possess latent reasoning capability that training signals fail to unlock?
- Does latent reasoning capability exist in base models before any training?
- Can minimal training signals unlock reasoning already latent in pretrained representations?
- Does the base model already contain latent reasoning capability?
- Can models reason at inference without specialized internal training?
- Do base models truly possess latent reasoning capability?
- Can targeted activation steering surface latent reasoning in base models?
- What other triggers can activate the latent reasoning capability?
- Why does latent reasoning override no-think instructions in models?
- Can pretraining signals unlock latent reasoning that post-training merely activates?
- What makes reasoning capability a pre-training rather than post-training phenomenon?
- Can activation-space steering vectors replicate thinking model performance without retraining?
- What makes some reasoning strategies genuinely novel versus latent?
- What is the distinction between teaching reasoning how versus when to activate?
- Can RL training teach models when to activate reasoning versus when to skip it?
- What distinguishes reasoning activation mechanisms across different training methods?
- How much training data is truly necessary to unlock latent model reasoning?
- Can distillation from stronger models create genuinely new reasoning abilities?
- Can latent reasoning architectures work as retrofits to existing models?
- Can instruction-level interventions fix memory-induced reasoning failures in practice?
- Can activation steering compress reasoning without retraining models?
- Can we predict when a model will develop thinking behaviors?
- Can we steer model reasoning by manipulating single features?
- Can structured workflows unlock latent reasoning abilities that raw models don't show?
- How do two-phase training dynamics explain reasoning emergence?
- Can models learn when to invoke search during reasoning tasks?
- Can activation steering vectors compress reasoning without retraining models?
- Does this reasoning steering method work consistently across all model sizes?
- Why does inference-time thinking hurt proactive critical thinking in vanilla models?
- Do base models already contain latent behavioral principles waiting to be amplified?
- When does knowledge activation fail across different model architectures?
- What makes thought identifiability provable without auxiliary training data?
- Does RL training activate latent meta-learning capacity or create it from scratch?
- How does policy initialization with sub-policies enable emergent thinking?
- Can activation steering directly steer models toward concise reasoning without prompting?
- Why does pre-training provide the raw material for emergent thinking?
- Can auxiliary modules preserve reasoning without catastrophic forgetting?
- What other latent LLM capabilities remain inactive without explicit activation cuing?
- Does RL training actually restore the critical thinking that reasoning models lose?
- How does an instruction-following LLM activate latent retrieval knowledge?
- Does base model geometry predict which associations persist through intervention?
- Can you steer reasoning by directly manipulating SAE features?
- What makes some concepts more steerable than others in activation space?
- What makes a model fail to activate relevant skills from its own harness?
- How does the prefrontal cortex inspire artificial reasoning architectures?
- How much reasoning catalyst data is actually needed for improvement?