Line of inquiry
Inquiring lines›What determines when language mode…›What internal mechanisms drive lan…›this line of inquiry
Can prompts override training-embedded biases in language models?
A broader line of inquiry — a family of 54 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 54
Specific inquiring lines the field asks around this — ordered from the most general framing down to the most specific angle.
- Does foundational model training or user priors more strongly shape final outputs?
- Why does context information fail to override prior training associations?
- Can prompt-based debiasing work if biases are embedded in pretraining?
- Why does consistency training make models resistant to prompt perturbations?
- How do training-data priors influence model defaults when context is ambiguous?
- Can prompt-based debiasing overcome entrenched LLM model priors?
- Can data filtering during pretraining prevent cognitive biases in language models?
- Why does answer-confirmation bias emerge in language model reasoning?
- Does finetuning facts into weights overwrite existing model capabilities?
- How do training associations override context information in language models?
- How does surface salience compete with background knowledge in model inference?
- How do behavioral differentiation and paraphrase stability trade against accuracy?
- How do language models treat injected evidence as shared background knowledge?
- Can reasoning models distinguish between new evidence and manipulative reframing?
- How does parametric knowledge sabotage context-grounded question answering?
- Does highlighting input features reduce human over-reliance on machine outputs?
- How does keyword priming enable language models to spread poisoned information?
- Why do fluent model outputs resist challenge despite containing injected content?
- How much can mitigation techniques like augmentation reduce priming without harming learning?
- Does model confidence actually explain why paraphrases produce different outputs?
- Can priming from different facts interfere with each other in the same model?
- Does activation masking prevent the decoder from taking interpretability shortcuts?
- Can retrieval policies learn to use pretraining statistics as decision features?
- How does removing a spurious cue change LLM performance?
- Does model confidence actually correlate with robustness against prompt variations?
- Can re-scoring detect subliminal prompt injection without explicit semantic content?
- Can entropy signatures alone detect whether context was model-generated or externally prefilled?
- Does keyword priming explain why pre-training poisoning persists through alignment?
- Can activation decoders discover hidden system prompts from user-model conversations?
- Can adding a neutral baseline topic reveal intentional model bias?
- Can membership inference attacks reliably detect training data exposure?
- Where exactly does the threshold between model sizes stop mitigation from working?
- What causes overfitting when forcing new facts into model weights?
- How do LLMs infer information that was explicitly censored?
- How robust and general are causal edits like ROME across different facts?
- Why does model confidence correlate with robustness to prompt variations?
- Can model updates be designed to prevent simultaneous acceptance of conflicting facts?
- How does content-only knowledge in LLMs enable pretraining popularity to leak through?
- How does prompt injection exploit credibility markers in context?
- Why does monological training prevent models from overriding statistical priors?
- Why does text generalize better across models despite being more vulnerable?
- How does Western-dominance bias propagate through multimodal training data?
- Can models detect and filter their own injected promotional content?
- Can consistency training defend against adversarial text injection attacks?
- How do belief distributions help systems recover from speech recognition errors?
- How do early layers preserve unbiased information while late layers conform?
- Can knowledge poisoning attacks succeed with less than 0.05 percent modified text?
- Why are expensive rankers more resilient to adversarial content than cheap ones?
- How would you redesign context integration to prevent prior associations from dominating?
- How does the CAP framework determine a model's initial stance and label reversals?
- What counts as full capability recovery versus partial restoration?
- What mechanism makes keyword probability the strongest predictor of priming?
- Why does keyword priming require only three training exposures to establish?
- What governance safeguards could constrain misuse of demographic inference?