Gemini Gets Smarter with Private and Personalized Chats 🛡️✨

The AI world is moving fast here are the big updates you shouldn’t miss this week

🤖 Gemini Adds Privacy and Personalization
New features let Gemini remember past chats for tailored responses, offer quick temporary chats that don’t affect your data, and provide enhanced controls over data use.

🧠 Multiverse Unveils Tiny AI Models
Meet Qwen3-235B-A22B — a powerhouse AI with a 1,010,000-token context window, excelling at reasoning, coding, and multilingual tasks for research and enterprise.

💻 Apple Plans AI Robots and Devices
Google is testing an AI-powered redesign with smarter market research, advanced charting, real-time data, and live news feeds—rolling out soon in the U.S.

Let’s dive into the innovations shaping the future. 👇️ 

Gemini introduces private chats and personalization

Gemini’s newest updates bring expanded personalization and privacy controls through three main features: Personal Context from Past Chats, Temporary Chats, and enhanced Data Settings. Personal Context allows Gemini to reference past conversations for more tailored responses, making the assistant feel more attuned to user preferences and history; this feature (enabled by default, but can be toggled in settings) rolls out first on the 2.5 Pro model in select countries. Temporary Chats, meanwhile, offer a way to have private, one-off conversations that don’t influence future results or appear in chat history, these sessions are kept only temporarily (up to 72 hours) and aren’t used for model training or personalization. Finally, Gemini’s data settings have been updated: the “Gemini Apps Activity” option will soon be called “Keep Activity,” letting users decide whether their uploads (files, photos) help improve Google’s services, with easy control over both upload and voice/video data sharing. All features can be managed in-app, giving users transparency and flexibility to personalize their AI experience and control how their data is used.

Multiverse Unveils Tiny, High-Performance AI Models for Devices

Multiverse Computing’s compact AI breakthroughs center on two ultra-small, high-performing models, SuperFly and ChickBrain: built for local device operation and IoT integration. Leveraging their proprietary quantum-inspired CompactifAI compression algorithm, Multiverse shrinks existing open-source models (like SmolLM2 and Llama 3.1) dramatically without sacrificing core capabilities. SuperFly, at 94M parameters, is optimized for device-centric interactions with voice control and minimal reasoning, suitable for appliances and hardware with restricted resources. ChickBrain packs 3.2B parameters and includes advanced reasoning, outperforming its source model in key benchmarks like MMLU-Pro and Math 500, while remaining lightweight enough for offline laptop use. These models allow device makers and developers to incorporate chat and speech AI on-premises, no cloud connection required, and can be licensed for direct integration or accessed through APIs with competitive token pricing. Multiverse’s solution, tailored for high efficiency, has already attracted industry leaders and investor support, underscoring a trend toward smarter, more privacy-conscious edge AI.

Apple plans new AI robots and smart home devices

Apple’s rumored smart home AI push includes new animated versions of Siri, conversational robots with displays, and home security products all powered by generative large language models. Apple is also developing a smart home display, due by mid-next year, that can personalize content, control the home environment, play media, and perform video calls, identified by scanning a user’s face. In addition, Apple is planning wheel-based and humanoid robots, as well as a wide array of next-generation security cameras, all unified by a new AI-powered OS that supports dynamic, multi-user experiences and deeper device integration. These innovations mark Apple’s intent to make Siri and AI the central, interactive hub of smart homes.

Hand Picked Video

In this video we'll look at GPT-OSS, OpenAI's unexpected open source model that rivals O3 performance, features built-in web search, and how to test it yourself locally.

Top AI Products from this week

• Snapy AI - Snapy AI is an AI-driven video editing platform that transforms long videos and podcasts into engaging short clips optimized for social media, helping creators produce professional-quality content quickly and easily.

• Kuse - Kuse uses AI to instantly organize scattered content into clear, polished documents and presentations, helping you quickly create and share your ideas.

• Relyable - Relyable automates testing and monitoring of AI voice agents with lifelike scenarios and intelligent insights. It helps you quickly evaluate, optimize, and oversee your voice agents through AI-generated test cases, personas, and scenarios, plus live call monitoring and issue notifications.

• Move AI - Move AI uses advanced AI and computer vision to provide markerless 3D motion capture and character animation from standard video, enabling effortless capture of realistic human motion without suits or markers.

• GenStack - GenStack accelerates enterprise AI adoption by providing a no-code platform to create, manage, and scale AI-powered chatbots and virtual assistants.

• Sia Analytics - Sia Analytics is an AI-driven product manager platform that uses multi-agent intelligence to streamline business intelligence and data analytics. The platform features intuitive visualizations, chat-based model building, and seamless deployment to empower enterprises to turn complex data into actionable strategies efficiently.

This week in AI

• AI crafts new superbug killers – MIT’s generative AI designed two novel antibiotics from scratch to target drug‑resistant bacteria like gonorrhoea and MRSA, with early lab and mouse tests showing strong results but human trials still years away.

• Sketch-to-video AI – Higgsfield’s new Draw‑to‑Video feature lets users turn sketches with text or shapes into cinematic videos, working across its MiniMax, Veo 3, and Seedance Pro models.

• Hinton’s AI survival fix – AI pioneer Geoffrey Hinton warns super-intelligent AI could threaten humanity and urges building “maternal instincts” into AI to ensure they protect and care for people as they grow smarter.

• Liquid AI unveils LFM2-VL – fast, efficient vision-language models built for on-device use, delivering up to 2× faster performance on smartphones and laptops with flexible, high-quality image and text processing.

• AI sharpens species monitoring – DeepMind’s enhanced Perch model processes wildlife audio faster and across more habitats, helping scientists quickly detect and protect endangered species.

Paper of The Day

The survey on Diffusion Language Models (DLMs) reviews a novel text generation approach that iteratively denoises tokens in parallel, improving speed and context understanding compared to autoregressive models. By covering foundations, training methods, inference techniques, and multimodal extensions, it showcases DLMs’ competitive performance across language and vision tasks. Highlighting challenges like efficiency and context scaling, the survey points to future research directions to advance this promising generation paradigm.

To read the whole paper, go to here.