Identity-Based Personalization

Are truly personalized #AI assistants finally here? Imagine having your own AI assistant that truly understands your preferences, writing style, and habits. Sounds great, right? But the current approaches to personalization have some major drawbacks: 1. Fine-tuning a unique model for each user is too expensive and resource-intensive for widespread adoption. 2. Retrieving relevant user history as demonstrations can break the continuity and fail to capture overall patterns. A new model called PPlug (Persona-Plug) aims to solve these challenges by introducing a lightweight "user embedder" module. Here's how it works: 1. PPlug constructs a unique embedding (a numerical representation) for each user by analyzing all of their past interactions and contexts. 2. This lightweight user embedding is then attached to the input whenever that user makes a request to the AI system. 3. By considering this user-specific context, the language model can better understand the user's habits, preferences, and communication style to generate more personalized and relevant outputs. Experiments on the LaMP benchmark show that PPlug significantly outperforms existing personalized #LLM approaches - all without fine-tuning the base model. While there are still open questions around data privacy and scalability, this might be an important step towards making AI assistants feel more like personal companions rather than one-size-fits-all tools. arXiv: https://lnkd.in/du-bVa_K

How do we balance AI personalization with the privacy fundamental of data minimization? Data minimization is a hallmark of privacy, we should collect only what is absolutely necessary and discard it as soon as possible. However, the goal of creating the most powerful, personalized AI experience seems fundamentally at odds with this principle. Why? Because personalization thrives on data. The more an AI knows about your preferences, habits, and even your unique writing style, the more it can tailor its responses and solutions to your specific needs. Imagine an AI assistant that knows not just what tasks you do at work, but how you like your coffee, what music you listen to on the commute, and what content you consume to stay informed. This level of personalization would really please the user. But achieving this means AI systems would need to collect and analyze vast amounts of personal data, potentially compromising user privacy and contradicting the fundamental of data minimization. I have to admit even as a privacy evangelist, I like personalization. I love that my car tries to guess where I am going when I click on navigation and it's 3 choices are usually right. For those playing at home, I live a boring life, it's 3 choices are usually, My son's school, Our Church, or the soccer field where my son plays. So how do we solve this conflict? AI personalization isn't going anywhere, so how do we maintain privacy? Here are some thoughts: 1) Federated Learning: Instead of storing data in centralized servers, federated learning trains AI algorithms locally on your device. This approach allows AI to learn from user data without the data ever leaving your device, thus aligning more closely with data minimization principles. 2) Differential Privacy: By adding statistical noise to user data, differential privacy ensures that individual data points cannot be identified, even while still contributing to the accuracy of AI models. While this might limit some level of personalization, it offers a compromise that enhances user trust. 3) On-Device Processing: AI could be built to process and store personalized data directly on user devices rather than cloud servers. This ensures that data is retained by the user and not a third party. 4) User-Controlled Data Sharing: Implementing systems where users have more granular control over what data they share and when can give people a stronger sense of security without diluting the AI's effectiveness. Imagine toggling data preferences as easily as you would app permissions. But, most importantly, don't forget about Transparency! Clearly communicate with your users and obtain consent when needed. So how do y'all think we can strike this proper balance?

What’s in a name? More than you might think, especially for AI. Whenever I introduce myself, people often start speaking French to me, even though my French is très basic. It turns out that AI systems do something similar: Large language models infer cultural identity from names, shaping their responses based on presumed backgrounds. But is this helpful personalization or a reinforcement of stereotypes? In our latest paper, we explored this question by testing DeepSeek, Llama, Aya, Mistral-Nemo, and GPT-4o-mini on how they associate names with cultural identities. We analysed 900 names from 30 cultures and found strong assumptions baked into AI responses: some cultures were overrepresented, while others barely registered. For example, a name like "Jun" often triggered Japan-related responses, while "Carlos" was linked primarily to Mexico, even though these names exist in multiple countries. Meanwhile, names from places like Ireland led to more generic answers, suggesting weaker associations in the training data. This has real implications for AI fairness: How should AI systems personalize without stereotyping? Should they adapt at all based on a name? Work with some of my favourite researchers: Siddhesh Pawar, Arnav Arora, and Isabelle Augenstein Read the full paper here: https://lnkd.in/e-WMukjQ

Stop getting it wrong with #AI: So, here’s the thing, if your idea of AI personalization is slapping a customers name on a promotional email or serving up “Customers who bought this also bought…” pop-ups, then congratulations—you’re stuck in 2012. I came across a solid research published in #HBR by #BCG on how leaders and laggers in various industries are applying AI. Yes consultants have a habit of putting things into frameworks and metrics, but this one was good BCG. Breaking some #myths on #Personalization and #AI : • 🧟Myth 1: AI is just for automation. No, it’s not. AI is for making people feel like you get them. #Netflix doesn’t just automate recommendations—it fine-tunes them to your weirdly specific taste for crime dramas, maybe some dark content with a hint of comedy. That’s connection. • 🧟Myth 2: Personalization = profits. In reality, loyalty and trust bring growth and add to profits. #Starbucks tailors offers through its Rewards app, focusing on loyalty first—and the profits follow. • 🧟Myth 3: Data hoarding equals success. Spoiler alert: it doesn’t. Collecting data without actionable insights is like hoarding junk. #Amazon, on the other hand, uses its data so well that 35% of its revenue comes from its AI-powered recommendation engine. It integrates browsing habits, past purchases, and customer reviews to suggest items that resonate. To quantify personalization maturity index multiply the below metrics: 1️⃣Empower Me(50%) Personalization starts with solving real problems not just offering flashy features. Example #Alibaba’s AI-driven tools empower small businesses by providing tailored logistics and financing solutions. 2️⃣Know Me(10%) Understanding your customer is essential. #Sephora’s AI-driven app uses purchase history and skin tone matching to suggest relevant products. 3️⃣Reach Me(10%) Timing and channels make or break personalization. #Uber’s predictive AI sends ride prompts exactly when users are most likely to need a car ride. Contrast this with brands that bombard customers with irrelevant offers, eroding trust. 4️⃣Show Me(10%) Visual and contextual relevance elevate personalization. #Sephora’s virtual try-ons demonstrate how personalized content enhances decision-making. Companies that rely on generic or mismatched ads lose credibility & engagement. 5️⃣Delight Me(10%) Creating unexpected moments of joy: #Spotify’s “discover weekly” doesn’t just predict your mood but it surprises and delights customer with a 56% engagement rate to prove it. 6️⃣Remaining 10% score weightage is attributed to CXOs championing AI projects Companies that treat AI-powered personalization as a strategic imperative, rather than a cost-cutting tool, stand to gain the most. Leaders like Netflix, Uber, Amazon, Starbucks, Spotify, Alibaba Group and SEPHORA dominate the Personalization Maturity Index because they’re masters of combining AI with human-centric strategies. Meanwhile, laggers just don’t know how to turn data into meaningful actions.

AI Personalization, Privacy, and Pleasure: Getting the Trade-offs Right View My Portfolio. Personalization can boost comfort and adherence—but only when privacy, consent, and safety are engineered first. A practical framework for intimate AI • Data minimization: Collect the least necessary. Defaults: no account, local-only logs, clear toggles to disable. • On-device first: Run recommendations on-device where possible; sync only anonymized aggregates for opt-in users. • Transparent value exchange: Tell users exactly what they get (e.g., “better comfort settings in 2–3 sessions”) and what is never collected. • Consent as a workflow: Plain-language prompts at setup, re-consent after major updates, and one-tap data resets. • Guardrails: Hard caps on intensity/temperature, cooldowns to reduce receptor fatigue, and safe-word style stops for interactive modes. • Explainability: Replace “AI decided” with “We noticed you prefer lower frequency after 3 minutes—would you like to save this?” • Bias checks: Test across anatomy, age, and sensitivity ranges; track who benefits and who doesn’t, then adjust. • Secure by default: Encrypted storage, ephemeral session data, and short retention windows; no shadow analytics. Metrics that matter • Adherence: ≥60% 30-day follow-through without pushy notifications. • Comfort: ≥4/5 median comfort after 10 minutes of guided use. • Personalization lift: +15–25% fewer manual adjustments after week 2. • Privacy trust: <1% opt-out due to data concerns, measured via voluntary in-app survey. At V For Vibes, we curate and develop products that treat personalization as a clinical-grade feature: local-first intelligence, explicit consent, and transparent coaching—so users get better outcomes without sacrificing privacy. #SexTech #SexualWellness #AI #Personalization #PrivacyByDesign #HumanFactors #InclusiveDesign #DigitalHealth #ProductAnalytics #TrustAndSafety #VForVibes

Can we create AI experiences that feel personal without feeling invasive? In today's digital age, balancing privacy and personalization in AI products is more crucial than ever. As we navigate the complexities of user trust and data usage, it’s clear that how we handle user data can make or break the success of AI-driven solutions like Copilot, ChatGPT, Gemini, etc. Here are a few key insights to shape consumer AI products that respect privacy while enhancing user experiences. 📍 Clear Communication on User Data - Distinguish between personally identifiable information (PII) and anonymized data. - Simplify communication to prevent user overwhelm and privacy concerns. 📍 Balanced Transparency - Find the right level of openness that reassures users without causing unnecessary alarm. - Focus on key data usage aspects that directly impact users. 📍 Non-Intrusive Personalization - Enhance user experiences without feeling invasive. - Avoid over-personalization that feels creepy. 📍 Modular Identities and Privacy Controls - Recognize and accommodate users' multifaceted personas. - Provide flexible privacy settings to manage different identities. 📍 Reducing Cognitive Load - Simplify privacy controls to reduce decision fatigue. - Focus on essential controls that are easy to navigate and understand. So as designers how do we make the experience better? ✅ Clarity in Communication - Keep privacy communication concise and clear. - Regularly update users on data use policies. ✅ Empower with Context - Use contextual prompts and just-in-time privacy notifications. - Reinforce users' control over their information. ✅ Value-Driven Personalization - Ensure personalization is contextually relevant and immediately valuable. - Communicate the tangible benefits of data usage. ✅ Chooser-Directed Experiences - Drive personalization by user consent and control. - Provide clear options for customization and easy revocation of consents. ✅ Embrace Modular Identity - Design flexible privacy settings for varying degrees of openness. - Accommodate users' diverse privacy needs across different life aspects. ✅ Simplify Privacy Settings - Prioritize simplicity to reduce cognitive load. - Use intuitive mechanisms like sliders for easy privacy management. ✅ Progressive Disclosure - Start with an overview and invite users to explore detailed explanations. - Ensure transparency without overwhelming users with information. The paradox of personalization vs. privacy is real. As we strive to balance these trade-offs, ensuring Chooser privacy is a fundamental aspect of the user experience. By innovating responsibly and embracing a user-centric approach, we can lead in technology while upholding ethical AI product-making standards. #ai #dataprivacy #personalization #uxdesign #aidesign #designthinking #copilot #chatgpt #generativeai #dataprotection #aiethics #inclusivedesign

AI shopping is having an identity crisis. It knows what you bought last Tuesday, but still can’t show you how you'd actually look in that outfit. It’s personalized… just not for you. That’s where Glance AI stands out. In a world of algorithms recommending algorithms, Glance makes product choices that feel surprisingly human. I’ve been playing around with their platform recently, and honestly? They’re doing something different. Take a selfie, and suddenly you’re seeing yourself styled in real-time with hyper-realistic visualization. Not product photos. Not mood boards. You. Actually you. This isn’t just better personalization. It’s identity-first commerce. While Pinterest just upgraded their visual search and Visa launched “Intelligent Commerce” for AI agents, most platforms still make you imagine yourself in their stuff. Glance AI flips that completely - they put you in the center and build the experience around your actual face, your actual style, your actual aesthetic language. What Shopify did for sellers, Glance AI is doing for shoppers. And Mary Meeker’s latest report confirms we’re all feeling it - AI adoption is happening at “unprecedented” pace, but most of it still feels like better algorithms serving the same old experience. Here’s what has me thinking: For creatives: This eliminates the mood board-to-reality gap entirely. You’re not guessing what something might look like - you’re seeing it on you, in real-time. For storytellers: Instead of generic personas, you’re working with AI that speaks individual aesthetic languages. It understands your visual identity. For the industry: Glance AI has integrated its proprietary models with cutting-edge platforms including Google Gemini and Imagen on Vertex AI delivering hyper-realistic, personalized experiences to users. But here’s the bigger shift: We’re moving from catalog-centered to identity-driven commerce. From “Here’s what we have” to “Here’s who you could be.” What if shopping started with you instead of the store? This is early, but it’s pointing toward something fundamental: AI that doesn’t just recommend products, but shows you possibilities. Not what the algorithm thinks you want, but what you might become. The question isn’t whether AI commerce will replace traditional e-commerce. It’s where this kind of identity-amplification actually takes root. What patterns are you seeing in AI that puts humans first instead of products first? If you’re interested in checking out Glance AI, take a look here: Apple App Store (iOS): https://lnkd.in/etGqg6Qt Google Play Store (Android): https://lnkd.in/e25EF6xi Glance #GlanceAI #AICommerce #FutureOfCommerce #ad

🤩 What if all your LLM interactions could be used to create a personalized layer that continuously evolves to adapt to your style? You wouldn’t need to keep reminding the model to match your style—it would just adapt naturally over time. Since we’re already giving LLMs so much data, why not use it to create a user-specific layer? Here's some good work in this direction! ⛳ A new paper proposes "PPlug", a lightweight plugin that creates a user-specific embedding based on historical user behaviors, allowing LLMs to tailor outputs without altering their structure. The plugin operates on a plug-and-play basis, meaning it improves personalization without retraining the model. It captures holistic user behavior rather than focusing on specific instances (generally done in RAG like approaches), leading to better adaptation to user preferences. It doesn't treat all user data equally, it selects relevant historical behaviors and synthesizes them into a personal embedding, based on their importance to the current task. 🤔 I really feel like personalization isn’t getting the attention it deserves, even though we have the technology to do it well. Probably because it’s quite expensive at this point, but it’s definitely something that will gain traction soon. Just imagine LLM-based personalization integrated into all AI products—it would completely change how we interact with tech. Link: https://lnkd.in/egWA_UZK

Infinite-ID Identity-preserved Personalization via ID-semantics Decoupling Paradigm Drawing on recent advancements in diffusion models for text-to-image generation, identity-preserved personalization has made significant progress in accurately capturing specific identities with just a single reference image. However, existing methods primarily integrate reference images within the text embedding space, leading to a complex entanglement of image and text information, which poses challenges for preserving both identity fidelity and semantic consistency. To tackle this challenge, we propose Infinite-ID, an ID-semantics decoupling paradigm for identity-preserved personalization. Specifically, we introduce identity-enhanced training, incorporating an additional image cross-attention module to capture sufficient ID information while deactivating the original text cross-attention module of the diffusion model. This ensures that the image stream faithfully represents the identity provided by the reference image while mitigating interference from textual input. Additionally, we introduce a feature interaction mechanism that combines a mixed attention module with an AdaIN-mean operation to seamlessly merge the two streams. This mechanism not only enhances the fidelity of identity and semantic consistency but also enables convenient control over the styles of the generated images. Extensive experimental results on both raw photo generation and style image generation demonstrate the superior performance of our proposed method.

Memory & personalization might be the real moat for AI we’ve been looking for. But where that moat forms is still up for grabs: •App level •Model level •OS level •Enterprise level Each has very different dynamics. 🧵 ⸻ 1. App-level personalization Apps build their own memory & context for users. Examples: •Harvey remembering firm-specific legal knowledge for law firms •Abridge capturing patient conversations & generating notes for doctors •Perplexity building long-term search profiles for individual users ➡️ Most likely in vertical applications with focused use cases and domain-specific data. This is where Eniac Ventures is currently doing most of our investing ⸻ 2. Model-level personalization The model itself becomes personalized and portable across apps. Examples: •ChatGPT memory & custom instructions •Meta’s LLaMa fine-tuned on personal embeddings ➡️ Most likely in general-purpose assistants and broad horizontal use cases where user context needs to travel across apps. ⸻ 3. OS-level personalization Personalization happens at the OS level, shared across apps & devices. Examples: •Google Gemini native to Android •Apple (maybe) embedding Claude via Anthropic ➡️ Most likely in consumer devices and mobile ecosystems where platforms control distribution. ⸻ 4. Enterprise-level personalization Each enterprise owns and controls its own personalization layer for employees & customers. Examples: •Microsoft Copilot trained on company data •OSS models (LLaMa, Mistral) deployed on private infra with platforms like TrueFoundry •OpenAI GPTs fine-tuned & hosted in secure enterprise environments ➡️ Most likely in highly regulated industries (healthcare, financial services) where data privacy and compliance are critical. ⸻ Why it matters: Where memory & personalization “land” may define who captures AI value. Different layers may win in different sectors. Where AI memory lives may reshape who captures the next decade of value.