personalization

Microsoft Research 介绍 StudentSim 研究系统，通过基于公开去标识化学习记录训练的个性化 AI 学生模拟器，在真实学习者参与之前评估并改进 AI 辅导工具，同时发布跨国际象棋、英语写作和数学的评估基准 StudentSimEval。

A Wired review of AlterMe's DNA-powered fitness and wellness program, which combines a genetic cheek-swab test, a smart ring, and an app for personalized workouts, nutrition, and recovery. The reviewer found the habit-building ecosystem motivating but noted premium pricing and limited revelatory value from DNA insights.

The paper proposes pFedDPO, a personalized federated direct preference optimization method where a shared base model is synchronized via FedAvg while private LoRA adapters capture client-specific preferences, reducing gradient diversity heterogeneity compared to vanilla FedDPO.

Orbora 是一款通过 Product Hunt 发布的 AI 音频产品，可为用户生成个性化的引导式音频，包括冥想、呼吸练习、ASMR 和短篇智慧分享，支持自定义语音、时长、语气、背景音和主题，并可离线收听。

本文提出 Diffu-LoRA，一种通过可学习门控与双层优化实现非均匀秩分配的参数高效方法，用于个性化文本到图像扩散模型，在 DreamBooth 等数据集上相较基线提升了主体保真度与提示对齐效果。

开源 Chrome 扩展 AI Homepage 正式公开源码，它用 Anthropic API 代理读取用户的浏览历史和页面内容，自动生成可自定义风格的个性化新标签页主页。

A software engineer's blog post criticizing the rise of LLM-generated 'personalized' recruiter emails that praise a candidate's background but are blatantly disconnected from the actual job description, making outreach feel even more impersonal than old templated spam.

提出了 PersonTTS，一个通过需求匹配的控制器初始化和源蒸馏程序化指导，实现个性化测试时缩放（多维用户需求联合满足）的摊销式智能体策略发现框架，在 AIME 和 HMMT 上显著优于现有 TTS 基线。

This Samsung Semiconductor paper proposes LINEUP, which decouples personalization from per-user adaptation by learning a bank of reusable low-rank factors and restricting each user to only 8 trainable scalars instead of 4.19M per-user LoRA parameters, outperforming baselines on all 12 metrics across six personalized LLM tasks.

The paper proposes Approximate Pareto Optimality (APO), a method for collaboratively learning lightweight LLM aligner initializations that support few-shot personalization under data deficiency, grouping users by compatible updates and coordinating competing objectives toward Pareto-optimal points with proven suboptimality bounds.

A creator built a custom Skill for their AI agent that surfaces local events, pop-ups, and experiences happening around New York City, helping the agent understand the user's physical environment rather than just internet content. They're sharing it to gather feedback and see if others would find it useful for their own agents.

Nikita Bier argues that AI enables any product to be designed and manufactured exactly to a consumer's specifications, citing his own experience using AI to design a custom dog door with a wifi-controlled lock that was fabricated and delivered for near mass-produced prices.

The author speculates that LLMs trained on vast surveillance and social-media data could predict individual user behavior and deliver hyper-personalized persuasion at scale, raising concerns about autonomy, preference manipulation, and invisible discrimination.

The author shares a personal AI news aggregation pipeline using MinHash/LSH for near-duplicate detection and JEV (plus openjev) for role-based classification, outlining future plans for relevance, impact, reliability, and accuracy signals feeding a personalized daily email digest.

This arXiv preprint develops a theory of context sufficiency for generative AI personalization, proposing a Context-Sufficiency Frontier that distinguishes four states (insufficiency, sufficiency, saturation, interference) and showing via a full-factorial experiment with a generative recommender at a home-furnishing retailer that relevant context improves appropriateness while irrelevant context harms it and destabilizes retrieval.

This paper formalizes personalized preference learning as a geometry-aligned optimization problem and proposes GAP-DPO, an iterative algorithm that performs utility-aware, gradient-aligned preference pair selection, showing improved stylistic fidelity and preference alignment over standard DPO variants.

I built newsroom, a small open-source app written in Clojure on jolt that reads RSS feeds and web searches, then uses an LLM to write a personalized daily news briefing with source-linked claims, scheduled runs, and incremental updates across days.

A user reported that Google's Gemini correctly revealed their mother's name without ever being told it, then offered three conflicting explanations — from denied knowledge to background context to regional statistics. The likeliest cause is Gemini's integration with connected Google services like Gmail and Contacts, so the post advises checking personalization settings before assuming an account compromise.

An opinion piece arguing that AI models are getting smarter but still fail to retain context across sessions, making 'agent intuition' — learning from past conversations and adapting to changing preferences — the next frontier in human-AI interaction.

OuterSignal announced a $22M Series A to expand its agentic customer intelligence platform, which uses trained internal models to reveal detailed customer profiles for e-commerce and consumer brands.