Large language models (LLMs) are increasingly used in long-running user-facing applications where responses may need to maintain a consistent interaction style. Prompting is easy to reuse but often weak, while fine-tuning and representation interventions can be stronger but are typically model-specific. We present PeLA, Prompt-Enhanced Logit Adaptation, a parameter-efficient framework for Big Five personality-trait steering. PeLA keeps the base LLM frozen, uses a personality prompt to specify the desired trait direction, and applies a learned logit adapter during decoding. The adapter contributes a base-relative residual controlled by a token-position gate, refining the prompt-conditioned distribution only when intervention is useful. Because the control module operates in output-token space, it can be reused across models with identical tokenizer vocabularies and token-id order. On open-ended PersonalityBench questions, PeLA improves macro-average target-direction scores over a strong Personality Prompt baseline on Qwen3-8B and Qwen2.5-7B, while cross-size experiments demonstrate tokenizer-compatible transfer within both the Qwen3 and Llama-3.2 families after validation-based scale calibration. PeLA narrows the gap to model-specific steering while preserving a frozen base model.

Pre-trained language models (PLMs) leverage chains-of-thought (CoT) to simulate human reasoning and inference processes, achieving proficient performance in multi-hop QA. However, a gap persists between PLMs’ reasoning abilities and those of humans when tackling complex problems. Psychological studies suggest a vital connection between explicit information in passages and human prior knowledge during reading. Nevertheless, current research has given insufficient attention to linking input passages and PLMs’ pre-training-based knowledge from the perspective of human cognition studies. In this study, we introduce a Prompting Explicit and Implicit knowledge (PEI) framework, which uses prompts to connect explicit and implicit knowledge, aligning with human reading process for multi-hop QA. We consider the input passages as explicit knowledge, employing them to elicit implicit knowledge through unified prompt reasoning. Furthermore, our model incorporates type-specific reasoning via prompts, a form of implicit knowledge. Experimental results show that PEI performs comparably to the state-of-the-art on HotpotQA. Ablation studies confirm the efficacy of our model in bridging and integrating explicit and implicit knowledge.