Tag
Xiaohongshu (RedNote) 研究团队发布 Vela 框架，将真实用户行为记录转化为训练任务与轨迹，通过监督微调和强化学习提升 LLM 在用户相关任务（+11.5 分）及通用基准（平均 +4 分）上的表现。
Vela提出以真实用户行为数据作为监督信号，将10万用户的匿名行为记录转化为可训练任务与轨迹，用于微调Qwen3.6-35B-A3B，在用户相关任务上最高提升11.5分，通用基准平均提升4分，表明真实用户行为是可扩展的监督来源。
Anthropic 更新使用政策，禁止对 Claude 模型的辱骂行为，主要执法手段是结束对话而非封禁账户，但政策违规仍可能面临限流、暂停或终止账号等处罚。
Google is integrating AI into its search engine through features like AI Mode and AI Overviews, which has resulted in millions of users engaging with AI, often unintentionally, and transforming online behavior and trust in AI responses.
A WIRED article explores the growing trend of people using AI chatbots to create custom fiction, citing a study that finds over a third of ChatGPT interactions involve fiction generation.
The article highlights how users often underutilize the iPhone 18 Pro Max's advanced features, sticking to basic tasks, and introduces 11 hidden features to unlock its full potential.
Splitsense uses AI to analyze user behavior and increase conversion rates for businesses.
The AI Observatory project provides independent analysis of real AI conversations, revealing that actual usage differs significantly from data reported by companies like Anthropic and OpenAI, highlighting more diverse and sensitive behaviors.
This paper introduces LUNAR, a benchmark for evaluating how large language models personalize responses from longitudinal app interaction histories across daily-life domains such as clothing, food, housing, and mobility. Experiments on 19 mainstream LLMs reveal that effective personalization depends on evidence selection and cross-domain integration, and that stronger personalization can come at the cost of privacy protection.
Kubit launches product analytics aimed at optimizing AI agent actions by analyzing user behavior.
LangChain highlights IO-HMM from GetCandidly, a design that separates user behavior (observable signals) from agent behavior (controllable inputs) in conversation turns.
The developer of Thumio announces that the next three upgrades are based purely on behaviors observed among top users, highlighting a data-driven approach to feature development.
The article explores, in video format, how users gradually become addicted to AI through three feedback loops: cognitive decline, hyper-productivity demands, and emotional loneliness. It distinguishes three levels of use: tool, agent, and authority, warning that AI may quietly take over human decision-making.
An article discussing the growing public fascination with watching AI agents perform tasks, and the question of what to call this phenomenon.
A study from Incogni reveals that 55% of Americans have stopped posting on social media, driven by digital fatigue and privacy concerns. The article explores the shift away from public sharing and the role of data brokers.
This article explores whether AI has altered how people search for information, comparing AI-driven search tools with the traditional use of Google.
A reflection on building a support bot reveals that users often do not search for help, highlighting key behavioral patterns for AI support design.
Elman believes that AI has fundamentally changed what ordinary users can do, and Gen Alpha, as native world-builders growing up with Roblox and Minecraft, have no preconceptions about how apps should be used. Together, these two things open a new window of opportunity.
This paper proposes NaviGen, a framework for personalized multimodal content generation that encodes user behavior into executable instructions using a dual identifier and a two-stage SFT+RL pipeline, improving personalization across product, game, and short-video domains.
Strange search queries often contain valuable product signals rather than being mere noise, revealing user expectations, supply gaps, navigation issues, and regional needs. This insight is particularly relevant for AI agents, where queries can initiate operations, making query analysis a strategic product concern.