Ezra Eeman’s Post

"How People Use ChatGPT" by the National Bureau of Economic Research (https://lnkd.in/efn4WEHy) makes for very interesting reading. However I am not sure how to interpret the data because  Usage ≠ Incidence. Why our ChatGPT metrics need both? I found the recent ChatGPT usage analysis genuinely enlightening, but it reports message volume (“usage”) rather than unique sessions (“incidence/instances”). That distinction matters. When I open ChatGPT once (one incidence) I may need several back-and-forth turns to land the right answer in the right context (multiple messages = usage). If we only measure usage, we may be counting all those clarification loops as “more use,” when they can just as easily indicate friction: unclear prompts, ambiguous outputs, or missing context. However that could be the undeclared intentions by the authors (?) The point is this is more than semantic and needs clarity. It reminds me of a consulting assignment I did years ago with financial services client to (re)segment the market. The meeting was intended go on for an hour but we took most of that hour debating what is 'an account'.

What do people actually use ChatGPT for? It’s a question that has lingered since the tool first went viral back in 2022. Now, a new research paper from OpenAI sheds light on user behavior by analyzing a sample of 1.1 million messages from active ChatGPT users between May 2024 to July 2025. The findings, summarised in a helpful visualisation by Made Visual Daily, show that ChatGPT’s core appeal is utility: helping users solve real-world problems, write better, and find information fast. Source: Visual Capitalist https://lnkd.in/gcjw-7ty

OpenAI launched ChatGPT Pulse, and it's a bigger shift than people realize. Instead of waiting for you to ask questions, ChatGPT now researches overnight and delivers personalized morning briefings. It learns from your chats, connects to your calendar and email, then proactively surfaces what you need to know. This flips AI from reactive to proactive. For 3 years, we've been stuck asking AI questions. Now AI is anticipating what we need before we even think to ask. What this means for business: → No more empty dashboards waiting for input → CRM systems that research prospects before meetings → Security tools that spot threats before they happen → Marketing platforms that adjust campaigns in real-time We just crossed the line from "AI as a tool" to "AI as a thinking partner." The companies that get this shift—and build proactive AI into their workflows—will have a massive advantage over those still playing catch-up with reactive AI. Are you ready for AI that works while you sleep? https://lnkd.in/gUE-FqiA

If they don’t understand it, they won’t buy it. That’s why I love the new ELI10 (Explain Like I’m 10) feature in ChatGPT. It’s not just a fun gimmick. It’s a reminder: Simplicity wins. Especially in sales. Most clients don’t care about vector embeddings or context windows. They care about: ✅ “Will this save me time?” ✅ “Will this reduce my costs?” ✅ “Will this help my team perform better?” You don’t earn a YES by sounding smart. You earn it by speaking their language, solving their pain, and showing clear ROI. In AI consulting, we’ve learned: If you want clients to trust automation, start by removing confusion. Next time you pitch your product, ask yourself: Can I explain this like I’m talking to a 10-year-old?

ChatGPT moment of 2022 inspired the world to rethink what is possible. For me, it was making enterprise data simpler and we started building an AI-powered query engine. Easier said than done. In our very first demo, we couldn’t even answer: "Show total sales in North America." Fast forward to September 2025: Tursio can now handle the most complex, open-ended questions. In fact, we challenge our users to find a better answer. So what changed? We systematically reduced ambiguity for LLMs by ensuring: - grounding in metadata - coherence with semantics - guidance by valid examples - consistency with query processing rules - and as close to the user intent as possible Fun fact: the response below ends with “straightforward consolidation”, overlooking the blood, sweat, and yes, the tears, behind it :)

OpenAI’s new ChatGPT Pulse offers subscribers personalized, AI-curated daily updates and research briefs each morning, synthesizing past chats, calendar events, and feedback. Designed to shift from reactive Q&A to proactive assistance, Pulse aims to help users start the day informed and organized https://lnkd.in/ewDGTgQU

🚀 Exciting news from OpenAI: ChatGPT Pulse is rolling out to Pro users on mobile! Instead of waiting for your prompts, Pulse proactively researches and delivers personalized daily updates in clear, scannable cards you can expand for more detail. Even better — you can shape what it brings by giving feedback or connecting apps like Gmail and Google Calendar. As Fidji Simo shared (https://lnkd.in/drHNMbht), this marks a big shift: ChatGPT is moving from simply answering questions to anticipating needs — helping you stay ahead and move your goals forward. This is just the first step toward an AI that works alongside you, thinking in real-time about how to help you succeed. 👉 Learn more: https://lnkd.in/dakMvDU8 #OpenAI #ChatGPT #Pulse #AI #ArtificialIntelligence #Productivity #FutureOfWork #Innovation #AIAssistant #GenerativeAI #WorkSmarter #AIForBusiness #DigitalTransformation

Everyone thought ChatGPT would make great copy effortless. Instead, it made inboxes… 10x noisier and 10x easier to ignore. Now? → Everyone’s email sounds the same → Everyone’s using the same “frameworks” → Everyone forgot how to sound like a real fkn person In 2025, the new edge isn’t automation. It’s tone. → Quirky beats polished → Spiky beats neutral → Vivid beats “valuable” AI can help you write faster. But only you can write with guts.

💡 How much of the world’s data does ChatGPT actually know? If the world’s data equals 100%, ChatGPT has trained on less than 1% of it. Yet from that tiny slice, it can generate insights, write code, explain complex ideas, and assist across industries. The power isn’t in “knowing everything,” but in how it compresses patterns from data and helps humans amplify their thinking.

🔍 You turned off ChatGPT data sharing. Good move. You can sleep soundly. But what do you think happens when you click that little 👍 or 👎? You might imagine something simple: - A short summary sent somewhere? - Maybe the last exchange reviewed by a human team to improve answers? Probably… but not only that. A quick read through the Terms of Use reveals something interesting: when you provide feedback, "the entire conversation may be shared for model training". If that’s what you intended - contributing to improve the model by sharing the whole feed- then all good. But if your chat includes personal or sensitive info…you might want to think twice before rating a response. 💬 What’s your take: should platforms make this more explicit, or is it on us to read the fine print? [source: https://lnkd.in/etvVyeX5]