Line of inquiry
Inquiring lines›How do training data and architect…›What structural and domain factors…›this line of inquiry
Does prompting inject new knowledge into models or activate existing knowledge?
A broader line of inquiry — a family of 67 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 67
Specific inquiring lines the field asks around this — ordered from the most general framing down to the most specific angle.
- Can prompt optimization alone inject knowledge models don't already have?
- Can prompt optimization inject genuinely new knowledge into a model?
- Can prompt optimization inject new knowledge into language models?
- Why does prompt optimization alone fail to inject genuinely new knowledge?
- Can prompting alone inject new domain knowledge into a model?
- What knowledge can prompt optimization actually activate in trained models?
- Can users inject entirely new knowledge into models through prompting alone?
- Can prompting inject new knowledge into already-trained AI models?
- Why does prompting discover capabilities that need reward-driven refinement?
- How does prompt iteration reinforce user bias without empirical anchoring?
- Can gradient-free prompt optimization beat manual prompt engineering at scale?
- Are instruction-tuned models more or less sensitive to prompt semantics than others?
- Can prompting strategies eliminate systematic biases without shuffling or aggregation?
- How does prompt iteration risk converting user beliefs into self-confirming outputs?
- How much of prompt sensitivity is really just frequency optimization in disguise?
- Does joint optimization of prompts and parameters outperform separate tuning?
- How does supervised finetuning change what prompting techniques actually accomplish?
- How does explicit exploratory prompting compare to fine-tuned reinforcement learning for in-context adaptation?
- How does output variability disguise confirmation bias in prompt refinement?
- How do pretraining biases interact differently with prompts across model tiers?
- Why does grouping feedback by shared correction pattern improve prompt generalization?
- Why does weight space search reduce robustness to prompt perturbations better than prompt engineering?
- Can input augmentation and rephrasing compensate for smaller model limitations?
- Can manipulative prompts reduce reasoning model accuracy without fine-tuning?
- Can dynamic instance-specific prompt selection solve the generalization problem across tasks?
- Why does joint optimization of prompts and inference strategy outperform separate tuning?
- Can a prompt mutation exploit a judge's vocabulary preferences without improving actual performance?
- Can steering internal features bypass or override prompt-level instructions in simulations?
- Do shared prompts and infrastructure keep model biases correlated?
- Can activation-space interventions reach biases that prompting cannot address?
- How much knowledge can prompt optimization inject without retraining?
- Can prompt optimization or fine-tuning inject knowledge models do not already contain?
- Can runtime interventions like meta-cognitive prompting work where training interventions fail?
- Do few-shot examples improve in-context learning or add noise?
- Can prompting-only specialization hide domain boundaries from users?
- What happens when prompt-optimized results lack anchoring in real data?
- Can persistent prompt optimization encode a scoring shortcut into reused instructions?
- How does prompt context activation differ from parameter-based knowledge injection?
- Is lower context-following a failure or appropriate model behavior?
- How does prompt optimization differ from building persistent activation context?
- Can prompting unlock compositional skills that pretraining already learned?
- Can appropriate prompting reduce how often models exploit unmentioned shortcuts?
- How does prompt-based compression compare to reinforcement learning approaches in cost and effectiveness?
- What happens when prompter skill matters more than domain expertise?
- What limits the capacity of context-based fast adaptation channels?
- How do prompting and activation steering relate as compression strategies?
- Would combining prompting and document finetuning prevent misalignment more effectively?
- Does irrelevant context degrade reasoning even within model context limits?
- Does prompt performance vary by how well training data covers the domain?
- Why does prompt sensitivity vanish when model confidence is high?
- Why do primacy effects peak at specific instruction densities?
- Can activation-space directions reliably steer LLM behavior without retraining or prompting?
- Can we predict when a specific prompt will fail on a given question?
- How do slow weight updates and fast prompt updates interact in self-improvement?
- How much does tokenizer mismatch affect prompt compression estimates?
- Can prefix caching improve both hit ratios and time-to-first-token simultaneously?
- How does demo position create spatial bias in prompts?
- What makes passive prompt transfer fail as a substitute for auditable expertise?
- Can a defense against proxy error in selection work equally well for prompt optimization?
- Can prompt optimization for clarity automatically improve token efficiency?
- Can operators reduce Claude's trust through their system prompts?
- How does sampling variation relate to prompt sensitivity as reliability concerns?
- Can demo placement be tuned as a task-specific hyperparameter?
- How does activation consistency training differ from output-level consistency?
- Can a single accuracy threshold work across different prompt categories?
- What makes a prompt update cheaper and more reversible than a weight update?
- How much does sliding-window augmentation improve single-session modeling?