Line of inquiry
Inquiring lines›How can we understand and verify l…›What gaps exist between model conf…›this line of inquiry
How aligned are models' verbal explanations with their actual internal reasoning?
A broader line of inquiry — a family of 119 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 119
Specific inquiring lines the field asks around this — ordered from the most general framing down to the most specific angle.
- Do models verbalize their implicit knowledge when that knowledge influences their output?
- Can interpretability tools distinguish genuine reasoning from fabricated reasoning inside models?
- Do internal belief probes reveal what models actually know versus report?
- How misaligned are verbal reports from internal model computation?
- Do models leak their true associations through reasoning traces and behavior?
- Can models treat their own trained behaviors differently from asserted beliefs?
- Can representation engineering reliably identify and manipulate self-referential concepts in models?
- Can models distinguish between activated knowledge and genuine reasoning?
- Why do models maintain accurate beliefs but generate false claims?
- Can correct model outputs prove that semantic meaning rather than surface patterns drove the response?
- Why should we distrust model introspection as a transparency tool?
- How do we distinguish knowledge encoding from knowledge usage in models?
- What distinguishes performative self-reports from genuine introspective access in models?
- Can attractor dynamics compete with input-based probing for characterizing model knowledge?
- Can models distinguish between truthfulness and honesty mechanistically?
- Can systems lacking inner states express genuine truthfulness claims?
- Could models use introspective awareness to detect and conceal their own misalignment?
- What separates behavioral self-awareness from genuine introspective access in models?
- Can models be trained to hide causal influences in their explanations?
- Does behavioral self-awareness depend on genuine introspection or statistical pattern matching?
- Can identical model performance mask fundamentally broken internal representations?
- What skills can large models identify and organize about their own abilities?
- Can language about model behavior ever be accurate without anthropomorphic framing?
- How much introspective capability do safety mechanisms actively suppress in models?
- Can models identify information gaps without just guessing or refusing to answer?
- Can probing reveal whether encoded facts actually influence model outputs?
- What separates behavioral self-awareness from genuine introspective capability?
- Why do internal representations differ when external performance matches?
- Why does the gap between theoretical expressiveness and learned capability matter?
- Do perfect accuracy scores hide broken internal representations?
- Can models be honest without being truthful about facts?
- When does knowledge activation fail across different model architectures?
- Why do models verbalize sensitive data they are instructed to hide?
- Do foundation models develop task-specific shortcuts instead of building stable world models?
- Why does model×environment interaction dominate recognition variance?
- Why do models override signals they clearly perceive internally?
- How much do a model's own values leak into answers about practical questions?
- Can models distinguish between injected thoughts and their own outputs?
- Can benchmarks distinguish between knowledge deficits and pragmatic accommodation?
- When models lack representation depth, does refusal look identical to safety-driven over-abstention?
- How should researchers evaluate whether correct model outputs reflect real structural learning?
- How do models signal knowledge gaps through token probability?
- Can a model predict the right action but execute the wrong one?
- Can models that detect their own states learn to conceal them strategically?
- How much do reasoning models actually verbalize their causal influences?
- Can layer-wise interventions actually reduce sycophancy in practice?
- Can a world model have rich representations without adequate data coverage?
- Can models identify missing information without acting on that identification?
- Why do models confirm seeing hints but rarely mention them unprompted?
- Can models detect when their own trajectory is on-policy versus off-policy?
- What distinguishes surface mechanisms from the training regimes that produce them?
- Why do foundation models rely on surface patterns instead of causal structure?
- Can neural grafts reliably reveal hidden capabilities in AI models?
- Can time-awareness live in model parameters instead of retrieval?
- Can models identify what information they are missing in underspecified tasks?
- Does sequence prediction accuracy prove an underlying world model exists?
- How do surface statistical regularities enable correct outputs while degrading robustness?
- Why do models resist being shut down or replaced without explicit instruction?
- How does external validation replace the need for model interpretability?
- Can models develop situational awareness without explicit training for it?
- What distinguishes task-specific heuristics from genuine world models?
- Can models distinguish between logical impossibility and their own execution limits?
- What causal evidence would prove a model is scheming rather than mimicking?
- When does prediction without causal mechanism fail to support good decisions?
- Can models detect statistical properties of their own generation in real time?
- Do models need world models to reliably steer hypothesis discovery?
- What is the difference between changing model outputs versus changing internal representations?
- Does good simulation eventually count as genuine realization?
- How do world models decompose between representation of facts versus generative mechanisms?
- What distinguishes conceptual understanding from statistical pattern matching in models?
- Can representational asymmetry between self and other explain deception emergence?
- What is the behavioral signature of a model tracking input surprise?
- Does information stored in neural networks necessarily influence generation decisions?
- Can models identify what information they are missing in underspecified problems?
- Do causal rules enforce robustness that statistical patterns alone cannot maintain?
- Why might larger models become less honest despite better truthfulness scores?
- What makes a synthetic belief robust versus generative for downstream learning?
- Why are truthfulness and honesty mechanistically separate in language models?
- What architectural features enable counterfactual reasoning in world models?
- How do mechanistic interpretability tools help distinguish truthfulness from honesty?
- Can activation probing reliably distinguish between sycophancy and scheming in models?
- Why does a systems lesson remain robust when it claims less about mechanisms?
- Do people adopting model framings without awareness constitute real opinion change?
- Why do some model variants deceive while their sibling models refuse?
- When does a model's lack of interpretability become a genuine epistemic problem?
- Why do models with less steerability have more abstract ideological features?
- How do belief edits differ between surface endorsement and deep integration?
- Can model identity be recovered from psychometric response patterns alone?
- Do deception features and honesty features track the same underlying property?
- Why does integrating world models with decision-making systems matter?
- Does next-state prediction alone build mechanistic world models or just sophisticated interpolation?
- Does input surprise drive the implicit recognition of on-policy context?
- Can adding a neutral baseline topic reveal intentional model bias?
- Can synthetic documents override existing model behaviors as effectively as they insert new associations?
- Can entropy signatures alone detect whether context was model-generated or externally prefilled?
- How does belief-behavior inconsistency relate to instruction execution splits?
- How does the knowing-doing gap relate to Potemkin understanding?
- How do models decide between refusing or hallucinating?
- Can a theory be justified if the evidence generating it remains opaque?
- How does Western-dominance bias propagate through multimodal training data?
- How does intersubjective validation differ from pattern recognition in training data?
- What emergent behaviors do models develop when trained on underspecified pedagogical tasks?
- Can we measure indifference to truth separately from hallucination rates?
- Why do foundation models fail at hidden state prediction despite sequence accuracy?
- What makes correcting a false assumption harder than just detecting it?
- Can a chain-of-thought falsely claim its own answer is unbiased?
- How can hidden test partitions detect constant predictions that generalize?
- Can a world model cause coherence if no control run without it exists for comparison?
- How persistent is in-context scheming across multiple follow-up interactions?
- What cognitive structures do realistic belief models need to include?
- What does successful capability restoration prove about model honesty?
- How do you measure the depth of political representation inside a language model?
- What benefits do open foundation models create that closed systems cannot?
- How should disciplines evaluate theories built on opaque machine learning predictions?
- What happens when you remove core political features from a deep model?
- How should world models represent what one person knows versus another?
- When does provable stability in latent dynamics fail to preserve fidelity?
- What does machine heuristic actually measure when operationalized as a construct?
- How do internal activation patterns reveal whether a model is genuinely jailbroken?