🚀Today’s Top AI Headlines:

1. DeepSeek Math V2 solves Putnam like a boss: DeepSeek has stunned the AI and mathematics communities with the release of DeepSeek-Math-V2, an open-source Mixture-of-Experts (MoE) math reasoner that outperforms both cutting-edge proprietary systems and top human competitors. The model achieved a 118/120 score on the Putnam exam, one of the hardest mathematics contests in the world, and successfully solved 5 out of 6 problems from the IMO 2025, a feat that places it at the very top of global mathematical reasoning performance. Unlike typical LLMs, DeepSeek-Math-V2 uses a generator–verifier framework, where one model attempts a solution while another checks and corrects the reasoning. This process enables iterative self-debugging, significantly reducing logical errors and hallucinations. Its design rivals Google’s private Gemini Deep Think, but with one major difference: DeepSeek’s model is fully open-source. This represents a major shift in the AI landscape. While companies like OpenAI and Google continue to release closed frontier models, DeepSeek is pushing high-performance, transparent alternatives into the public domain. Researchers say the release could democratize advanced reasoning and accelerate math, physics, and engineering breakthroughs. The model is now available on GitHub and major open ML hubs, giving students, engineers, and researchers access to one of the strongest math solvers ever built.

   Source: DeepSeek

   🤖 Robi: “GPT-5 studying math while DeepSeek’s already solving for x² + world domination.”

2. Perplexity adds memory + multi-calendar support: Perplexity is rolling out major upgrades to its productivity ecosystem, starting with a smarter email assistant that can now handle multiple calendars at once, a feature designed for users who juggle Gmail, Outlook, and enterprise accounts. The assistant understands context across inboxes, schedules meetings, drafts responses, identifies conflicts, and coordinates time blocks across personal and professional calendars simultaneously. The company also introduced persistent memory, one of its most significant AI enhancements yet. With this update, Perplexity’s assistant can remember user preferences, past conversations, writing tone, shopping habits, research interests, and frequently requested tasks. This allows the AI to carry context forward into future queries, reducing repetitive explanations and making responses more personalized and relevant. For example, if a user prefers concise emails, specific product categories, or certain research formats, Perplexity will automatically apply those preferences. The memory system is fully user-controlled, with the ability to view, edit, or delete stored items at any time. Combined, these updates push Perplexity further into the realm of a true personal AI, one capable of managing communications, schedules, and ongoing work with a deeper understanding of user needs. Analysts say persistent memory places Perplexity in direct competition with OpenAI’s Assistants, Google’s Gemini Projects, and Anthropic’s Claude Workflows.
   Source: Perplexity

   🤖 Robi: “Great, now my assistant remembers everything... including my 2016 crypto regrets.”

3. Alibaba unveils AI Glasses (but can they find your keys?): Alibaba has entered the AI wearable market with the launch of Quark AI Glasses, a smart eyewear lineup powered by its in-house Qwen large language models and Quark digital assistant. Priced starting at 1,899 yuan ($268), the glasses aim to bring advanced multimodal AI capabilities into everyday life at a mass-market price. The device combines a camera, microphone array, and lightweight display to deliver real-time translation, object recognition, scene description, and voice-activated assistance. Users can ask the glasses to summarize text, identify landmarks, provide navigation, or generate search results hands-free. The Qwen models also enable hands-free emailing, note-taking, and contextual Q&A based on what the wearer is looking at. Early reviewers in China say the glasses feel like a blend of Meta Ray-Ban AI and Google’s early Glass prototypes, but with stronger AI reasoning and more natural language responses. The product is being framed as an everyday assistant designed for travelers, students, and professionals who want lightweight, on-the-go intelligence. Alibaba’s move positions it as a major competitor in the emerging AI wearables race, going head-to-head with Meta, Apple’s Vision products, and startups building agentic AR devices. The company plans to expand availability after initial domestic demand is evaluated.
   Source: Ali Zila

   🤖 Robi: “Step 1: See the future. Step 2: Still bump into a lamppost.”

🔍Beyond the Headlines:

1. OpenAI confirms API leak via Mixpanel breach: OpenAI has confirmed that analytics provider Mixpanel suffered a security breach exposing some API users’ names, emails, locations, and device information. No sensitive data such as API keys, billing details, or project content was compromised, but OpenAI warns affected users may face targeted phishing risks. The company has removed Mixpanel from all systems, initiated an internal review, and is sending notifications to impacted developers. Security experts say that while the breach is limited, it highlights growing risks associated with third-party analytics tools used by AI companies managing sensitive enterprise workloads.
   Source: OpenAI

   🤖 Robi: “Ah yes, the “your data was safe… except it wasn’t” industry standard.’’

2. NVIDIA’s ToolOrchestra trains smaller, smarter AIs: NVIDIA and the University of Hong Kong have introduced ToolOrchestra, a system suggesting that future AI gains may come from smarter coordination rather than bigger models. The system trains an orchestrator that decides when to reason internally and when to call specialized external tools. An 8B model using ToolOrchestra beat GPT-5 and Claude Opus 4.1 on Humanity’s Last Exam with 37.1%, while being 2.5× more efficient and faster. Even with unfamiliar tools, the model adapted seamlessly, hinting at a future where small models outperform frontier systems through intelligent tool use.

   Source: Arxiv

   🤖Robi: “Finally, an AI that knows when to Google something instead of pretending it knows.”

🤖Prompt of the Day:

Corporate Data Quality Management Plan

🤖AI Tools You Didn’t Know You Needed:

Problem: Meetings are a time sink, afterward you spend precious minutes (or hours) on manual note-taking, summarizing, extracting action items, and organizing follow-ups.

AI Solution: ReadMeeting automatically captures your screen + audio during meetings, then uses AI to transcribe and summarize the discussion, extracting key moments, decisions, and action items so nothing falls through the cracks.

AI Tool: ReadMeeting is an AI-powered meeting assistant that records meetings, transcribes and analyzes them, then delivers structured summaries, highlights, action items, and downloadable reports (e.g. PDF), no manual note-taking needed.

Helpful Features

• One-click meeting capture (screen + audio)

• AI-generated summaries, key-moments detection & action-item extraction

• Exportable reports (PDF) and organized meeting archives

• Team & collaboration support (organize meetings by teams, share summaries internally)