Tag
Google has launched its experimental AI app Dreambeans in India, its first market outside the U.S., with support for 13 local languages, aiming to bring personalized AI-generated visual stories to a broader audience.
The paper introduces P-ContextBench, a scalable benchmark framework that builds synthetic, population-grounded personas embedded in fragmented mobile-app traces to evaluate LLMs on personal-context capabilities including temporal tracking, pattern inference, and constraint-aware action boundaries, finding SOTA models score below 60% on key tasks.
Meta Superintelligence Labs researchers formalize 'adaptive personalization under epistemic uncertainty' and introduce a multilingual benchmark revealing a recognition–generation gap in LLMs, where models default to exploit-first behavior despite appropriate strategy identification; targeted training on internal strategy representations narrows the generalization gap.
作者论述AI个人助理中「连接」与「信任」的区别：连接只是能力半径，信任是可验证的预测准确度，需要通过提议→确认/纠正→记忆化的产品闭环不断累积。文章还强调上下文无法被复制、隐私是感知深度的前提，并抛出关于知晓边界和集成顺序的开放讨论。
Google is updating Chrome's autofill with AI that can pull details like flight confirmations, tracking numbers, and addresses from users' Gmail to fill out online forms, rolling out to U.S. desktop, iOS, and Android users as part of its Personal Intelligence initiative.
Dalet (daletlabs) announced a $7M pre-seed raise for an AI-powered shopping app that generates a hyper-personalized feed of lifelike images containing real, in-app purchasable products, founded by researchers from Google, Runway, and MIT and backed by Lightspeed VP.
该论文提出 PERPL 框架，通过 LLM 模拟读者的相对偏好来学习个性化事实核查解释改写，在保持事实忠实性的同时优于更大的通用 LLM。
iPrefBench is a user-centric benchmark evaluating whether personal mobile-use agents can infer missing preferences from activity history or ask for clarification in an executable iOS environment with a simulated user. Evaluation of 7 frontier agents shows a large gap: high action completion (up to 91.6%) but low full task success (up to 47.7%), with agents especially struggling to request necessary ask-only information.
Meta的欧洲广告收入增长因欧盟《数字市场法》要求推出“低个性化广告”选项而显著放缓，欧盟地区增速落后于美加市场，监管压力正成为其广告业务的实际拖累。
Radaric 是一款 AI 驱动的信息流阅读器，可整合 RSS、Telegram、X 和 YouTube，并通过用户对内容的批准、拒绝与解释反馈来学习编辑口味，将重复报道聚合为单一故事。
The paper proposes Preference-PCA Pluralistic Learning (P3L), a framework that jointly learns trajectory embeddings and shared preference directions via PCA from anonymized crowd feedback, enabling steerable personal alignment and fast personalized reward learning for embodied control tasks without requiring annotator identities.
JustAI 宣布正式发布其全新的全栈营销平台，以统一的 Marketing Harness 驱动整个营销工作流，包含可自我优化的客户旅程、统一客户上下文以及基于自然语言提问的增长分析功能。
该论文提出一种基于约束上下文老虎机的个性化 LLM 路由框架，通过 primal-dual GLM-UCB 算法在线学习用户偏好，在最小化推理成本的同时以高概率保证响应质量满意度。
MyLens AI 推出面向大规模 1-1 实时演示的产品，能基于用户提供的原始素材为每位观众生成个性化实时演示，并在演示中回答问题、动态生成新幻灯片，还会提供每场会话的表现分析。
提出了 SkillGrain 框架，将用户反馈分解为可独立执行的原子规则，预测其可迁移性并只保留适用于未来请求的部分，通过强化学习联合优化技能归纳与条件生成，以实现可复用的个性化技能学习。
PersEvalBench introduces a benchmark testing whether LLM evaluators can reliably judge contextually appropriate use of user persona information across 403 multi-episode scenarios, finding that current evaluators—especially at detecting under-personalization—remain far behind human judgment (63.9% vs 94.4%).
本文提出对稀疏混合专家（MoE）大语言模型进行个性化，通过训练 crosscoder 学习可解释的个性化隐变量，并利用因果干预验证这些隐变量及其关联专家足以复现个性化行为，相比稠密模型降低了多义性并实现了局部可定位的控制。
CarryOn 提出一个双循环框架，在冻结的视觉-语言-动作（VLA）策略上加入持久化个性化规划与执行监督，显著降低安全违规率并提升故障恢复成功率。
Facebook 将推出 Your Algorithm 功能，允许用户通过自然语言和“感兴趣/不感兴趣”标签直接定制自己的信息流与 Reels 推荐，Meta 表示已在其排序系统中更广泛地使用 LLM 实现实时个性化。
CraveLens 是一款 Chrome 扩展，利用设备端 AI 识别视频及网页选定区域中的食物，并为用户生成个性化的 Swiggy 购物车以供确认。该扩展现已在 Product Hunt 上线，免费使用，基于 Google Gemma 构建。