Muazma Zahid’s Post

Big news from Microsoft 👀 For the first time ever, Microsoft 365 Copilot users will be able to choose between OpenAI’s GPT models and Anthropic’s Claude models. Most people won’t notice a big change just yet, but it’s a clear sign that more choice and flexibility are on the way. Most people know ChatGPT - powered by OpenAI’s GPT-5, which is now available in Microsoft Copilot. Just a few days ago, Microsoft also announced its partnership with Claude, a conversational AI developed by Anthropic, known for its advanced natural language capabilities. Here’s what nonprofits need to know: ✅ Limited rollout - Only available through Microsoft’s Frontier Program (preview) - Your IT admin has to switch it on in the M365 admin center ✅ Where Claude shows up - Researcher Agent → use it for strategy, sector trends, or funding analysis - Copilot Studio → Claude is now an option for building custom nonprofit tools ⚠️ The fine print - Preview only (not yet in Word, Excel, or PowerPoint) - Claude runs outside Microsoft’s environment, so data handling works a bit differently - Available in just two features for now 💭 What do you think? Would you switch between models depending on the task, or just stick with one? #Claude #Microsoft #Microsoft365Copilot #OpenAI #GPT5

Just built something seriously cool: I’ve powered AI assistant using the open-standard Model Context Protocol (MCP) — think of it as a “USB-C for AI” 🔌 — and it’s already doing real work on my server. Here’s the story: I created an MCP server to manage my Apache HTTP Server setup. Now, through plain English prompts, my assistant can: * list all available sites on the server * show which sites are enabled vs disabled * disable a site (e.g., my portfolio) * re-enable it again * restart Apache itself The sequence demonstrated is simple: 1) “List all available sites.” 2) “Disable my portfolio site.” 3) Test it — it’s indeed down. 4) “Enable my portfolio site.” 5) Test again — it’s back up. That’s the power of MCP: it bridges the gap between natural language, AI, and infrastructure. It doesn’t just answer — it acts. If you’re into automating operations, chatbots, dev-ops or AI + tooling workflows: this is something worth playing with. Happy to share more about how I set it up. Note: For this demonstration, I used OpenAI platform's chat which allows to connect mcp server. We can connect with anywhere we want (that's the power of mcp), Even with our Telegram, WhatsApp etc. To Learn, Refer https://lnkd.in/ghHy4fcR #AI #DevOps #MCP #InfrastructureAutomation #Apache #Tooling

Heard of OpenAI’s new drop in market? 𝗔𝗴𝗲𝗻𝘁𝗞𝗶𝘁 It’s no code platform, designed to help developers and enterprises build artificial intelligence (AI) agents from prototype to production in few hours It offers drag and drop canvas, embedded with 3 primary components:  1. 𝗔𝗴𝗲𝗻𝘁 𝗕𝘂𝗶𝗹𝗱𝗲𝗿 for creating and testing agent logic  2. 𝗖𝗵𝗮𝘁𝗸𝗶𝘁 for embedding customizable chat interfaces  3. 𝗘𝘃𝗮𝗹𝘀 for agents to measure performance and understand agent behavior through trace grading. Also, to add:  1. It’s a kit, built on top of OpenAI’s Responses API which already is used by hundreds of thousands of developers.  2. It integrates with OpenAI’s Connectors Registry. That means agents can safely connect to any tools/apps through an admin dashboard. 𝗢𝗯𝘀𝗲𝗿𝘃𝗮𝘁𝗶𝗼𝗻𝘀 𝗮𝗻𝗱 𝗢𝗽𝗶𝗻𝗶𝗼𝗻: 🎯It certainly raises the competitive bar for rest AI platforms which are in race to offer simplicity and autonomous agent building for complex scenarios. 🎯Over past decade, AI was limited to Data Science community and tech savvy people. Time indeed has arrived, when AI is accessible to everyone (well- democratized). 🎯Once, which was aspirational, has now become operational. On one side, thanks to all advancement and evolution in this space, and parallelly kudos to enterprises embracing it with open arms and responsive governance. #AI #Evolve #StraightfromExperience #ABB

AI hype is endless. But which courses are actually worth your time in 2025? Here’s a curated list - all free, beginner-friendly, and focused on LLMs, agents, and RAG. 1️⃣ LangChain for LLM App Dev (DeepLearning.AI) – Build LLM apps with memory, chains, document QA. https://lnkd.in/d85HcaFW 2️⃣ ChatGPT Prompt Engineering (DeepLearning.AI/OpenAI) – Learn prompt design directly from OpenAI engineers. https://lnkd.in/dpmzmXd9 3️⃣ Building & Evaluating Advanced RAG (DeepLearning.AI/LlamaIndex) – Retrieval tricks + evaluation triad. https://lnkd.in/d28Y7WxX 4️⃣ Data Agents (DeepLearning.AI/Snowflake) – Multi-agent orchestration and evaluation. https://lnkd.in/dS-3WEqU 5️⃣ Safe & Reliable AI (https://lnkd.in/dGuMkc_4) – Add guardrails to stop hallucinations/leaks. https://lnkd.in/dkEmZEWB 6️⃣ Generative AI for Beginners (Microsoft Learn) – 18-lesson series with demos + Azure/OpenAI. https://lnkd.in/dk-vzGVc 7️⃣ Intro to Generative AI (Google Cloud Skills Boost) – Short overview of GenAI + Vertex AI. https://lnkd.in/d4fUH84f Deleting “random YouTube noise” → you scale faster. Because learning AI is about structured steps, not hype. The best investment in AI isn’t GPUs. It’s structured learning.

The words to focus on here? "If you’re serious about building digital fluency, you need more than one lens." Just like reading literacy requires that you read more than one book, AI literacy is best when you explore more than one AI tool.

Graduated from Foundervine’s Pulse Programme and Google’s AI Essentials course - with a respectable 96% 🎉 Here are five AI tools and tricks I’ve found useful from my AI journey: 🪩 GEMS (GEMINI) & CUSTOM GPTs (CHATGPT) I used to think these were super complex & required coding. Turns out, they’re simple. Instead of copy-pasting context every time, you can create one with your tone, references & brand notes - and it remembers it for next time. ❍ BETTER PROMPTS = BETTER OUTCOMES Include: • Task • Context (like persona & format) • References Then keep evaluating and iterating. 🪩 NOTEBOOK LM Upload your materials (articles, sites, docs) and it creates mini quizzes, audio summaries or answers – with no hallucinations. ❍ PERPLEXITY Excellent for deeper, sourced web research. 🪩 HUMAN IN THE LOOP (ALWAYS) The most important thing for me: AI should be a tool, not the brain. Keep human oversight – read, analyse, and refine. It’s great for frameworks and refining drafts, but only YOU can bring the unique perspective, accuracy, and creativity. Are any of these new to you? Or am I just late to the party... 👀 😅 #PulseByFoundervine #FoundervinePulse #LearnWithGoogle

Solution for those who see GPT too pricy: you don't have to pay a $200 monthly subscription. The solution is to use API and pay per-request, as you need. I initially funded my account with $10 and used it for months, without limits. Now LLMs are became more expensive, and I can spend $3 per day. Vibe coding not included. 1. Register at platform.openai.com to access the API. 2. Verify your identity. OpenAI's platform uses Persona. RU passports are not eligible - ask your friends for help if you haven't secured another citizenship yet. 3. Now you have access to GPT-o3 and GPT-5 models. 4. Get a good chat client. I like Chinese LobeChat: https://lobechat.com and prefer a self-hosted installation. You can start with their official website, though I don't guarantee your data will be stored there forever. There's also LibreChat https://lnkd.in/eS3vAZkq and other projects. 5. Generate an API key on platform.openai.com and enter it in your client (e.g. LobeChat). 6. Enjoy! A large request to GPT-5 costs about 6 cents, and for powerful -pro versions it's about $1. Bonus: You can use Anthropic, DeepSeek, Llamas via Together.ai, etc., and compare different models. #LLM #GPT

How to design agents for relevant and focussed learnings with less data and more compute? Amazing context set by Edan Meyer in his video (https://lnkd.in/gdXZTDRe) Edan Meyer outlines three key areas of research that the field needs to focus on - areas often avoided in popular discourse - to achieve a good vision of AGI, which he defines as a general-purpose learning algorithm inspired by animal learning. 1️⃣ Continue Learning: This means the agent should never stop learning. While current models utilize fine-tuning (training on a new task after initial training) and in-context learning, these methods have significant caveats. When trying to fine-tune a model multiple times, researchers often encounter 🤔 "catastrophic forgetting" (the model forgetting what it previously knew) or a ☠️ "loss of plasticity" (the model slowly losing its ability to learn over time). Furthermore, information learned via in-context learning is only learned temporarily until that information exits the context window. 2️⃣ The Ability to Learn from a Single Experiential Stream of Data: This requirement is the exact opposite of how current Large Language Models (LLMs) are trained. Current LLMs are typically trained by randomly sampling from billions of disjointed segments of text (like passages from books 📚 , articles 📰, or transcripts 📜 ) and feeding them to the model in enormous batches. If a model only learns from disjointed sequences, it will never gain the ability to reason about how it should act over much longer time horizons. ⏲️ Without experiencing the world through a continual stream of observations, like humans do, concepts such as episodic memory 🤔 and temporally correlated causality 🧐 become meaningless. 3️⃣ Scaling with Compute (Designing algorithms where more compute always leads to better performance): While current methods have figured out how to scale models to hundreds of billions or trillions of parameters when using massive static data sets 🔢 , they have not figured out how to scale effectively in the continual single stream experiential learning setting. Current learning methods fail when training massive models without a massive static data set or when stopping the use of huge batches of data because they were designed with a "big data mindset from the ground up". The goal 🥅 is to develop methods that, when given more compute, can use that compute to extract more out of what they are experiencing moment to moment. 😇 My 2 cents : Hit back the basics of Reinforcement Learning and Deep Learning, we have the answers and few models in last few months are setting the stage where step-wise-breakdown-and-problem-solving is ingrained into the training methodology of model leading to solving his 3rd point.

💡 You’re just building API wrappers, not solving real problems. This is one of the most common criticisms AI engineers hear. But here’s the truth 👇 🤖 Training and deploying a model anywhere close to GPT-4o, GPT-4.1, Gemini Ultra, or DeepSeek is not just about writing code — it’s about handling massive infrastructure requirements. 📊 These models have tens or hundreds of billions of parameters, and serving them requires specialized GPU clusters 🖥️⚡ — not a single product server. 🏗️ Even if a company trains its own model, they won’t host it alongside their product backend. They’ll still deploy it on separate clusters and access it through… you guessed it — APIs 🔌. Exactly the same way we call OpenAI, Google, Anthropic, or Groq models today. So when people say “it’s just an API wrapper,” they miss the point: ✅ The real challenge is how you integrate these APIs into real workflows. ⚡ How you optimize latency, scale requests, manage caching, and ensure observability. 🚀 How you turn a raw model into a production-ready system that solves business problems. That’s the work AI engineers are doing. And it’s not “just a wrapper.” It’s the difference between a demo 🎭 and a product that works at scale 🌍. #ArtificialIntelligence #AI #MachineLearning #DeepLearning #GenerativeAI #AIEngineering #MLOps #AICommunity #AITools #FutureOfAI