Claude Can Now Remember Your Past Chats🧠💡

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

🤖 Claude Gets Personal
Tailor your AI experience with Claude’s new personalization tools: search past chats, set account-wide preferences, fine-tune project instructions, and switch between custom Styles.

🧠 Qwen3’s Mega Memory
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.

📈 Google Finance Levels Up
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. 👇️ 

Personalizing Claude: Chat, Preferences, Projects & Styles

Claude’s personalization features let you tailor the AI to your needs through four main tools: Searching Past Chats, Profile Preferences, Project Instructions, and Styles. The past chat search (rolling out to Max, Team, and Enterprise plans) lets Claude reference previous conversations for seamless continuity, with the option to disable it in settings or delete specific chats to exclude them. Profile preferences are account-wide settings where you describe your general approaches, terminology, scenarios, and communication style for Claude to consider in all responses. Project instructions apply only within a specific project (paid plans) and provide targeted context, workflows, requirements, and roles for that project. Styles adjust Claude’s tone and formatting, enabling you to switch between concise, explanatory, or custom styles. These features can be used individually or in combination for optimal, personalized interactions.

Qwen3-235B-A22B-Instruct-2507: Ultra-Long Context AI Model

Qwen3-235B-A22B-Instruct-2507 is a powerful, state-of-the-art causal language model from the Qwen series, available on Hugging Face. It boasts 235B parameters (with 22B activated), 94 layers, and a 262,144-token native context window—extendable up to an impressive 1,010,000 tokens for ultra-long text processing. Trained with both pretraining and post-training, the model excels at instruction following, logical reasoning, code generation, subjective tasks, and robust multilingual support. Thanks to advanced techniques like Dual Chunk Attention and MInference, Qwen3 processes extremely long contexts efficiently and accurately, leading its class on many benchmarks for knowledge, reasoning, coding, and alignment. It supports integration with frameworks and tools like Transformers, SGLang, vLLM, and is deployable on platforms with substantial GPU resources. For flexible applications and agentic use cases, Qwen-Agent provides streamlined tool-calling and customization. This model is open-source under Apache-2.0 and is recommended for both research and enterprise-grade AI development.

Google Tests AI-Powered Google Finance with Smarter Research & Real-Time Data

Google is testing a redesigned, AI-powered version of Google Finance that offers smarter research, advanced charting, and richer market data. Users can now ask detailed, complex finance questions and receive comprehensive AI-generated analyses, along with easy access to relevant web sources. The update introduces advanced charting tools—such as technical indicators and candlestick views—for deeper market visualization, plus expanded real-time data, including commodities and more cryptocurrencies. A live news feed delivers up-to-the-minute market headlines. The new experience will roll out over the coming weeks in the U.S. on google.com/finance, with the option to switch between the classic and updated designs.

Hand Picked Video

Olly.social is an AI-powered social media tool that automates commenting, post creation, summaries, and viral predictions to boost engagement, supports custom AI personalities, multiple platforms, popular AI models, and offers free and paid plans with real-time analytics.

Top AI Products from this week

• nFactorial AI - Simular is building the world’s first autonomous computers—systems that can use software like humans do. From browsing websites and filling out spreadsheets to running creative tools and managing workflows, Simular enables a new era of human.

• My Juno Health: AI Doctor - Fully personalized to empower you to live your best life, achieve your highest goals, and unlock your full potential. It's your partner in vitality with a personalized wellness coach in your pocket. Helps fuel your ambition and enhances your daily experience.

• SuperCraft - SuperCraft lets teams design physical products with natural language.

• Weave - Weave understands engineering work by combining LLMs and domain-specific machine learning. We tell you how much work is getting done, how good it is and how much is done by AI.

• Rid - We make selling easier than buying. Just send us photos. We handle the rest. We research your product, find markets for them, negotiate with buyers, answer questions, and handle delivery/pickup.

• Zpeakr - Master any language naturally through AI-powered conversation. Learn English, Spanish, French, and Italian with real-time feedback from AI tutors that challenge you, correct your mistakes, and make you speak — anytime, anywhere.

This week in AI

• Faster LLMs - Apple’s multi-token prediction speeds LLMs up to 5× for math/coding by predicting several tokens at once, with safeguards to keep output quality intact.

• China’s AI Biotech Surge - Chinese AI drug discovery firms like CSPC, XtalPi, and Earendil Labs land multibillion-dollar Big Pharma deals, boosting China’s role in global pharma innovation.

• GPT-4o Returns - OpenAI brought back GPT-4o for paid ChatGPT users after backlash over GPT-5, with fans preferring its friendlier tone and creative, more natural responses.

• MXFP4 Cuts OpenAI LLM Costs by 75% - OpenAI’s MXFP4 4-bit micro-scaling floating-point format massively reduces memory and compute needs, cutting inference costs by 75%, enabling faster, cheaper LLMs on less hardware.

• Grok 4 Free for All - xAI makes Grok 4 free worldwide. In Auto mode, complex queries auto-route to Grok 4, or pick “Expert” to always use it. Limited-time higher usage limits.

Paper of The Day

The TeamMedAgents framework embeds six proven teamwork principles—leadership, monitoring, orientation, shared models, closed-loop communication, and trust—into multi-agent LLM medical decision-making. By dynamically assembling specialist agents and adapting collaboration to task complexity, it improves diagnostic accuracy across eight medical benchmarks.

To read the whole paper, go to here.