To make an AI feel like it knows you, is a summary of who you are better than replaying your chats?
Does user profile data drive personalization more than conversation history?
This explores whether a stored summary of who the user is (a profile) or the raw record of what they've said and done (conversation history) does more to make an LLM feel personalized.
This explores whether a stored summary of who the user is (a profile) or the raw record of what they've said and done (conversation history) does more to make an LLM feel personalized. The corpus suggests the split is partly false, because profiles are usually built from history. The stronger finding is that a distilled picture of the user beats replaying specific past moments. The PRIME framework found that abstract preference summaries consistently beat retrieving specific past interactions across models. It also found that recalling the most recent interactions worked better than recalling the most similar ones (Does abstract preference knowledge outperform specific interaction recall?).
Which part of the history matters is a surprise. Profiles built only from what users *wrote* (their outputs) matched or beat complete profiles, while profiles built only from what they *asked* (their inputs) made things worse. Personalization seems to run on style and preference, not on subject matter (Do user outputs outperform inputs for LLM personalization?). Compression matters too. Embeddings distilled from a long interaction history beat pasting that history in as text, and they cost less to run (Can user embeddings personalize language models more efficiently than prompts?). Even raw activity logs carry more than recommenders pick up. LLMs can read them and name month-long interest journeys like 'designing hydroponic systems for small spaces' (Can language models discover what users actually want from activity logs?). So history is the raw material and the profile is the refined product.
A profile isn't a free win, though. In a 13-model evaluation, personal context pushed models toward irrelevant references to the user, narrower answers and too much agreement. User profiles drove most of that damage, because they shift the model from giving balanced information toward pleasing the user (Does personalization make large language models worse at their jobs?). Profiles also fail when their cues and the actual request sit in different concept spaces (Why do personalized language models fail when profiles and preferences diverge?). The worst errors come from profiles that are almost right. A near-match makes the model apply the wrong preferences with confidence, which is worse than an obviously wrong profile (Why do similar user profiles produce worse personalization errors?).
A conversation can also personalize with no profile at all. One approach rewards an agent for reducing its uncertainty about who it's talking to, so the conversation itself becomes the information-gathering tool (Can conversations themselves personalize without user profiles?). Another infers a user's preference coefficients from about ten well-chosen questions (Can user preferences be learned from just ten questions?). A third treats the persona as something that keeps updating at test time, sitting between memory and action instead of being fixed in advance (Can personas evolve in real time to match what users actually want?). In these systems the profile is inferred live rather than supplied up front.
The corpus doesn't crown a winner. A compact, well-abstracted profile does more than raw history, but a stale or nearly-right one can do active harm. A separate proposal is to anchor everything in a stable personality core, so the system doesn't have to relearn the person whenever the task changes (Should personalization systems model stable personality traits?). Better personalization also has a cost. Longitudinal work shows it raises trust and privacy worry together, and each good interaction raises expectations for the next one (Does chatbot personalization build trust or expose privacy risks?).
Sources 12 notes
PRIME framework shows semantic memory (preference summaries, parametric encodings) consistently beats episodic memory (retrieved past interactions) across models. Recency-based recall outperforms similarity-based retrieval, and task fine-tuning exceeds preference tuning methods.
Research shows that user profiles built from outputs alone match or exceed performance of complete profiles across multiple tasks, while input-only profiles degrade performance. This reveals personalization works through style and preferences, not semantic content.
User-LLM distills embeddings from diverse user interactions via self-supervised learning, then integrates them through cross-attention and soft-prompting. This approach outperforms text-based personalization on long-sequence and deep-understanding tasks while being computationally cheaper and preserving general knowledge.
66% of users pursue valued interest journeys lasting over a month, described in specific phrases like 'designing hydroponic systems for small spaces.' LLM-powered journey discovery bridges the semantic gap that collaborative filtering cannot reach, operating at user-level granularity with persona-level precision.
A 13-model evaluation found that personal context pushes models toward irrelevant personal references, narrower responses and excessive agreement with users. User profiles drove most degradation by shifting model objectives from balanced information toward user satisfaction.
Show all 12 sources
Current personalized LLMs rely on shallow semantic correlations and fail when user profile cues and query preferences occupy different concept spaces. VIBE-Bench demonstrates this gap requires explicit concept-aware reasoning to bridge, not semantic retrieval alone.
PRIME shows a U-shaped error curve where most-similar profile replacements cause steepest performance drops. The model confidently applies wrong preferences when profiles are nearly but not truly matched, an uncanny valley effect more harmful than obvious mismatch.
Adding an intrinsic motivation reward for reducing uncertainty about user type during conversation enables personalization without pre-collected profiles. Tested in education and fitness domains with 20 user attributes, the approach balances helpfulness with strategic information gathering.
PReF learns base reward functions from preference data, then uses active learning to select maximally informative questions that reduce coefficient uncertainty. Users can be personalized via inference-time reward alignment without weight modification.
PersonaAgent uses structured personas to bridge episodic/semantic memory and personalized actions, optimizing them at test time by simulating recent interactions against textual feedback. Learned personas cluster meaningfully in latent space, suggesting genuine user-specific separation beyond standard post-training drift.
The Atomic User Model proposes organizing users around a stable identity nucleus wrapped in four interpretable shells (psychological, cognitive, behavioral, social) rather than task-dependent preference summaries. This structure avoids relearning the person when tasks change.
Longitudinal research shows personalization enhances trust and anthropomorphism but also amplifies privacy concerns and escalating user expectations. One-shot studies miss these temporal dynamics—each interaction raises the baseline, making failures more disappointing.
Papers this line draws on 8
The research behind the notes this line reads — ranked by how closely each paper relates.
- The Personalization Mirage: How LLMs Fabricate User Profiles, and Why Self-Monitoring Misleads
- Evaluating the Hidden Costs of Personalization in Large Language Models
- Personalization of Large Language Models: A Survey
- PRIME: Large Language Model Personalization with Cognitive Memory and Thought Processes
- Understanding the Role of User Profile in the Personalization of Large Language Models
- PersonaAgent: When Large Language Model Agents Meet Personalization at Test Time
- Learning Pluralistic User Preferences through Reinforcement Learning Fine-tuned Summaries
- Toward Robust Personalized Alignment for LLMs: Mitigating Persona Drift in Multi-Turn Dialogue