<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Sanskar Gupta on Medium]]></title>
        <description><![CDATA[Stories by Sanskar Gupta on Medium]]></description>
        <link>https://medium.com/@ssanskarr?source=rss-e73cd299325c------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*xCY-XPDrKKc-uMZNMwq77g.jpeg</url>
            <title>Stories by Sanskar Gupta on Medium</title>
            <link>https://medium.com/@ssanskarr?source=rss-e73cd299325c------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Tue, 17 Nov 2026 02:44:46 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@ssanskarr/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Creative Is Abundant, Attention to Test It Is Not]]></title>
            <link>https://ssanskarr.medium.com/creative-is-abundant-attention-to-test-it-is-not-c2cffdff0117?source=rss-e73cd299325c------2</link>
            <guid isPermaLink="false">https://medium.com/p/c2cffdff0117</guid>
            <category><![CDATA[facebook-ads]]></category>
            <category><![CDATA[performance-marketing]]></category>
            <category><![CDATA[product-management]]></category>
            <category><![CDATA[marketing]]></category>
            <category><![CDATA[adtech]]></category>
            <dc:creator><![CDATA[Sanskar Gupta]]></dc:creator>
            <pubDate>Tue, 03 Nov 2026 07:09:35 GMT</pubDate>
            <atom:updated>2026-11-03T07:09:35.343Z</atom:updated>
            <content:encoded><![CDATA[<h4>How brands are using Creatives to run Performance Marketing?</h4><p>I work in product at Hawky.ai. The best part of my job is talking with and understanding the marketing of 60–70 brands up close, who constantly use creatives to boost their growth, and then building products around it.</p><p>Now-a-days, making an creative has become almost free. A new video, a new hook, a new face, a new offer. AI can produce a hundred of them in an afternoon. Something that once took a designer a full week now takes minutes.</p><p>You would think that changes everything. It does. Just not in the way most people expect. Because when making a creative costs nothing, making it stops being the hard part.</p><blockquote>The hard part now is something most brands don’t even realise they’re running out of: the attention it takes to find out which ads actually work.</blockquote><p>That’s where the real game is now. Let me explain.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*TvxB00KFTjG6F6xPlQcPcA.png" /></figure><h3>The creative became the targeting</h3><p>For years, a big part of marketing skill was deciding who should see an ad. Which age group, which cities, which interests. People were genuinely brilliant at it.</p><p>Today, the ad platforms’ own AI does most of that. And the way it decides who to show an ad to is, more and more, by reading the ad itself. What it says, who it seems to speak to, what it promises.</p><blockquote><strong>So the creative now carries the weight that targeting used to carry.</strong></blockquote><p>You can see this in how the field itself is changing. A new role, the creative strategist, has appeared almost out of nowhere, and most of the people in it used to be performance marketers, agency owners, or people working at agencies a couple of years ago. The work didn’t disappear. It moved into the creative.</p><p>In the end, I think marketing is becoming a game of creatives.</p><h3>Every creative is a question</h3><p>Here’s something most people outside marketing never see. Nobody knows in advance which creative will work. Not the founder, not the agency, not the best creative person in the room.</p><blockquote>The only way to find out is to show ads to real people and watch what they do.</blockquote><p>The way I like to think about it, every ad is a question you’re asking your customers. “Do you care more about price or about ingredients?” “Does a founder talking land better than a customer talking?” “Does this work for a first-time buyer, or only for someone who already knows us?”</p><p>And every answer costs money, because every person who sees an ad is someone the brand paid to reach. Testing runs on attention, and attention is bought.</p><h3>More ads, fewer answers</h3><p>This is where I think the whole story flips.</p><blockquote>AI made asking questions free. It didn’t make answers any cheaper.</blockquote><p>In any week, a brand can only reach so many people, and every ad running that week competes for the same pool of attention.</p><p>Imagine a restaurant with a hundred guests for the evening and a kitchen that has made two hundred dishes. Every guest gets a tiny spoonful of two of them. At the end of the night, how much do you actually know about which dish is best?</p><p>Almost nothing. One hungry guest can make an average dish look like a hit.</p><p>To genuinely tell a good ad apart from a slightly better one, you need tens of thousands of people to see each of them. Multiply that by a few hundred ads, and most brands simply don’t have that much attention to spend.</p><p>So the platform quietly decides for you. It pushes money towards whatever looks good in the first few hours, and most ads barely get seen. They don’t lose. They just never get a fair chance. And because ads go stale over time, some of them die before anyone learns whether they were any good.</p><h3>Rules decide who gets a fair shot (something i understood)</h3><p>Here’s where it gets interesting.</p><p>The brands I see doing really well haven’t slowed down on making ads. If anything, they make more than ever. Creative velocity is a big part of how they hit their numbers.</p><p>But they pair that velocity with something most brands don’t have: very strict rules.</p><p>Through a lot of testing, they’ve worked out exactly how a new ad gets tested, how long it runs, what it has to prove, and when it earns more budget. Their campaign structures are simple. They barely touch budget tweaks anymore. That’s just not where they play.</p><p>In these brands, I think the only mechanical part of marketing left is writing those rules and making sure every single ad follows them.</p><p>At first, this sounds like it contradicts everything above. It doesn’t. <strong>The rules are how these brands ration attention. Every new ad gets the same fair shot, against the same bar, before it gets scaled. Nothing wins just because it got lucky in the first few hours.</strong></p><blockquote>Velocity without rules is two hundred dishes and a hundred guests. Velocity with rules is a kitchen that knows exactly how every new dish gets tasted.</blockquote><h3>But, velocity of what?</h3><blockquote><strong>The velocity that works is velocity of ideas, not velocity of copies.</strong></blockquote><p>Two hundred ads that only change the background colour ask the same question two hundred times. Ten ads built on ten different ideas ask ten different questions, with the same attention.</p><p>This is also why I don’t think creative strategy is a rigid skill you can learn from a checklist. It’s very dynamic. Sometimes an ad works, and sometimes the same idea flops a month later. What you really get good at is having ideas, and judging which ones deserve to be asked.</p><p>The rules are the system. The ideas are the edge.</p><h3>Judgement is the new scarcity</h3><blockquote>AI made creatives abundant. Everything else in marketing is slowly arranging itself around that one fact.</blockquote><p>The brands that win from here, I believe, are the ones that treat attention as their scarcest resource, protect it with discipline, and spend it on real questions instead of near-copies.</p><p>It’s funny. The technology that made making creatives abundant might end up making judgement the most valuable skill in the room.</p><p><em>thoughts are mine, written via claude : )</em></p><p>—</p><p>A few things that led me here:</p><ul><li>Talking to hundreds of performance marketers over the last year!</li><li><a href="https://www.youtube.com/watch?v=eErcrQQHzg4">Start a business with no money</a>, Think School with Eric Ries: treating every bet as a hypothesis to test.</li><li><a href="https://www.ama.org/press-releases/press-release-from-the-journal-of-marketing-market-researchers-and-online-advertisers-are-a-b-tests-leading-you-astray-a-new-study-says-they-could-be/">Where A/B testing goes wrong</a>, Braun and Schwartz, Journal of Marketing: why a winning ad may just have been shown to an easier audience.</li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c2cffdff0117" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How I Build AI Products on Shifting Foundations]]></title>
            <link>https://ssanskarr.medium.com/how-i-build-ai-products-on-shifting-foundations-9eb016131018?source=rss-e73cd299325c------2</link>
            <guid isPermaLink="false">https://medium.com/p/9eb016131018</guid>
            <category><![CDATA[future]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[startup]]></category>
            <category><![CDATA[product-management]]></category>
            <dc:creator><![CDATA[Sanskar Gupta]]></dc:creator>
            <pubDate>Sat, 23 May 2026 05:12:07 GMT</pubDate>
            <atom:updated>2026-05-27T08:07:01.881Z</atom:updated>
            <content:encoded><![CDATA[<h4><strong>My approach to building durable product moats when foundation models refuse to stand still</strong></h4><p>I’ve been a PM for a while now, and something has fundamentally changed about how I think about my job. The model I’m building on top of today won’t be the model I’m building on top of in six months. And honestly, that one fact has reshaped almost every product decision I make.</p><p>I have realised how much my thought process has shifted. I am always thinking ahead now, always picturing what the product will look like in six months, when a new model probably drops in the market.</p><p>So I thought I would write about it.</p><p>This piece is about how I’ve adjusted. What I look for in a roadmap now, what I kill, what I double down on, and the test I run every feature through before it gets built.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*cCuPSUiKnPARPa9-HjmbxQ.png" /></figure><h3>What actually changed about the job</h3><p>I think most PMs haven’t fully internalized this yet — the labs building GPT, Claude, and Gemini aren’t your suppliers. They’re your competitors. Every few months they ship an update that absorbs features people spent a quarter building. I was watching a podcast, in which Sam Altman said it plainly: companies built solely on GPT-4 would be “steamrolled” by GPT-5. He said it almost casually, the way you’d warn someone about a wave they hadn’t noticed yet.</p><p>The wave is real. AdCreative.ai — the one I think about most, because it’s closest to the AdTech world I work in — scaled to roughly $40M ARR and got acquired for about 1x ARR. The market basically said: we don’t see a moat here worth paying for.</p><blockquote>In my view, the common thread isn’t bad execution. They executed fine. They just built features the foundation models eventually shipped natively.</blockquote><h3>The model is the least durable part of your product</h3><p>In my opinion, the model is the most volatile, most commoditizable, lowest-margin component in your stack.</p><p>Which means everything I build <em>around</em> the model can compound. And everything I build <em>to compensate for</em> the model’s current limitations will get deleted by the next release.</p><p>To me, this inverts the usual moat question. I don’t ask “what will the next model commoditize?” anymore. I ask: “which layer of my product is the model, and which layers can <em>never</em> be inside the model?” The answer to the second question is where the real work is.</p><h3>The test I run on every feature</h3><p>Chris Pedregal at Granola has a line I’ve quoted probably a hundred times: “Don’t spend any time working on problems that will go away.”</p><p>It sounds obvious. It is not obvious in practice. Because the problems that are about to go away often <em>feel</em> like the most urgent ones — the model can’t handle long context yet, hallucinates on certain queries, can’t reason through multi-step logic reliably. Engineering around these gaps feels productive. It looks like a moat.</p><blockquote><em>So now, for every feature, I ask one question: when the next major model lands, does this get better, get absorbed, or get worse?</em></blockquote><p>If it gets better — meaning model improvements lift our product — I greenlight it. If it gets absorbed — meaning the model now does this natively — I kill it. If it gets worse — meaning we engineered around a limitation that no longer exists — I should have killed it three months ago.</p><blockquote>Granola’s own example sticks with me. <strong>Their first version didn’t work for meetings longer than 30 minutes because OpenAI’s context window was tiny. </strong>The instinctive PM move would have been to engineer around it. They consciously didn’t. They predicted longer context was coming for free, and they were right.</blockquote><h3>What a 10x smarter model still won’t know</h3><p>Here’s the exercise I find most useful.</p><p>Picture Claude 6 or GPT-7. Assume it’s 10x cheaper and 2–3x smarter than today. <strong>Now ask: what part of my customer’s job does it <em>still</em> not know how to do?</strong></p><p>The answer, almost always, isn’t about intelligence. It’s about reach. A 10x smarter model still won’t have my customer’s last 90 days of campaign performance. It won’t know their internal definition of “good creative.” It doesn’t know which audiences underperformed last quarter and why. It has no operational trust with the account manager. None of this is in any model’s training data. Ever.</p><p>That, to me, is where the durable moat actually sits. Not in clever prompts. Not in elaborate agent scaffolding. In the connective tissue between the model and the customer’s actual operating reality — their data, their workflows, their accumulated context, their trust loop. Foundation labs structurally can’t build this, because building it requires being inside the customer’s perimeter.</p><p>This is also why I think performance marketing AI has an honest opening. The data lives behind ad-platform walls. The workflows are specific and load-bearing. The trust is hard-won. None of this lifts when GPT-6 ships.</p><h3>The questions I ask before I build</h3><p>Putting this all together, here’s the actual checklist that runs in my head before any feature goes into a sprint.</p><ol><li><strong>When the next model lands, does this get better, absorbed, or worse?</strong> If the answer isn’t “better,” I think hard before greenlighting.</li><li><strong>If OpenAI shipped this exact feature for free inside ChatGPT tomorrow, would we still exist?</strong> A brutal question. Most features fail it. The ones that pass tend to be the durable ones.</li><li><strong>Which part of this lives inside the model API call, and which part lives in our context, data, evals, or workflow surface?</strong> The model part is commodity. The rest is mine.</li><li><strong>What is a 10x smarter model still missing to do this well?</strong> That missing thing — usually proprietary context — is what I should be building, not the prompt around it.</li><li><strong>Is our roadmap driven by production failures we’ve actually seen, or by hypothetical model-improvement bets?</strong> When I catch my team in the latter, I pull us back.</li></ol><h3>Where I’ve landed</h3><p>I think being a PM on foundation models is a fundamentally different job than it was even three years ago. The product surface under you is moving. The competitive surface above you is moving. Both are moving faster than your release cycle.</p><p>What that means, to me, is that I’ve stopped optimizing for today’s model. I bet on capability curves instead of capability snapshots.</p><blockquote>I always assume the next model will be 10x cheaper and 2–3x smarter, and I ask whether my product is positioned to ride that or get crushed by it.</blockquote><p>The companies that figured this out — Cursor, Harvey, Midjourney, Granola — aren’t smarter than everyone else. They just made one bet that everyone else didn’t. They bet harder on the model improving than the model labs themselves did. And they built everything else — the IDE, the workflow, the community, the daily habit — to be lifted by that wave instead of drowned by it.</p><p>That, more than anything, is the shift I think every PM building on foundation models has to make. The model isn’t your product. What you build around it is.</p><p><em>thoughts and research are mine, written via claude : )</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9eb016131018" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Way India Shops Changed Quietly. The Consequences Are Anything But.]]></title>
            <link>https://ssanskarr.medium.com/the-way-india-shops-changed-quietly-the-consequences-are-anything-but-06c17da2020d?source=rss-e73cd299325c------2</link>
            <guid isPermaLink="false">https://medium.com/p/06c17da2020d</guid>
            <category><![CDATA[business-strategy]]></category>
            <category><![CDATA[d2c]]></category>
            <category><![CDATA[indian-business]]></category>
            <category><![CDATA[entrepreneurship]]></category>
            <category><![CDATA[indian-startups]]></category>
            <dc:creator><![CDATA[Sanskar Gupta]]></dc:creator>
            <pubDate>Sat, 28 Mar 2026 11:00:12 GMT</pubDate>
            <atom:updated>2026-03-28T11:00:12.443Z</atom:updated>
            <content:encoded><![CDATA[<h4>Why did HUL acquire Minimalist for ₹2700 Crore?</h4><p>If you are even a bit excited about the D2C space, this is going to be an interesting read!</p><blockquote>For decades, distribution created trust.</blockquote><p>In India, if a product showed up in every store, on every shelf, across every city, customers didn’t question it, they trusted it. Ubiquity itself was the signal. That’s how giants like Hindustan Unilever, ITC Limited, Marico, and Nestlé built empires. Get everywhere first. Trust follows.<br>Simple. Brutal. And it worked for 50 years.</p><p>But then something flipped.</p><blockquote>Today, trust creates distribution.</blockquote><p>If enough people trust a brand online, if Reddit threads are full of praise, if a dermatologist on YouTube swears by it, if repeat purchase rates are insanely high, platforms like Blinkit and Zepto don’t need convincing. They come to the brand. <strong>Consumers pull products into distribution. </strong>No one pushes them there.</p><blockquote><strong>And I think that one shift explains one of the strangest business patterns in India right now.</strong></blockquote><p>HUL spent ₹2,700 crores on Minimalist. ITC spent ₹500 crores on Yoga Bar. Marico bought four brands almost simultaneously — gourmet popcorn, plant protein, skincare, men’s grooming.</p><p>So what’s really going on?</p><blockquote>Why are companies with armies of MBAs and billion-dollar balance sheets paying thousands of crores to acquire brands that a couple of founders built out of a rented office?</blockquote><p>Let’s start from the beginning.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*lKhTUzm2Mp07PLR1TJm0uA.png" /></figure><h3>Why did this FMCG playbook that worked for 50 years suddenly stop working?</h3><p>For decades, the FMCG playbook in India was simple and brutal. Get your product onto every kirana store shelf in the country. Repeat. Protect it.</p><p>Companies like HUL, ITC, and Marico spent 30 to 50 years building the machinery to do exactly this — trucks, factories, distributors, salespeople, deep retailer relationships. Once you owned the physical shelf space in millions of kiranas, no new player could touch you. You won by being everywhere.</p><p>Then the shelf moved.</p><h3>Why is your phone screen now more valuable than a kirana shelf for brands?</h3><p>The way Indians discover and buy products has completely changed. Instead of walking into a store, <strong>people open Blinkit, Zepto, Amazon — or just scroll Instagram.</strong> A kirana shelf has maybe 200 products. An app has millions. But you can only see 8 to 10 products on your screen at once.</p><blockquote>That tiny screen became the new shelf. And winning on it requires a completely different playbook.</blockquote><p>The old system rewarded companies that could push products into stores at scale. The new system rewards brands that pull consumers toward them — through reviews, ratings, community, and word of mouth. <strong>When someone searches “niacinamide serum” at midnight, the product that wins isn’t the one with the best distribution. It’s the one with 10,000 five-star reviews and a Reddit thread where people swear by it.</strong></p><p>HUL’s Lakme doesn’t win that search. Minimalist does.</p><p>Here’s exactly how that shift played out — the old model on the left, the new one on the right:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*T5-jFVzIXVX15cLP8WzrIw.png" /></figure><p>The entire direction of flow reversed. And the FMCG giants — built entirely for the left side — suddenly found themselves unable to compete on the right.</p><p>Which brings up the obvious question. They have billions. Why not just build these brands themselves?</p><h3>Why couldn’t a billion-dollar company out-build a two-brother startup from Jaipur?</h3><p>HUL launched five of their own digital-first D2C brands. Poured resources in. Put their best people on it.</p><p>You know how much revenue all five brands generated combined?</p><p>Less than Minimalist — a brand two brothers started in Jaipur four years ago with zero FMCG experience.</p><p>That’s not a small miss. That’s a signal.</p><p>The issue isn’t capability. It’s identity. The categories where D2C brands are winning — skincare, protein, wellness — are ones where consumers deeply care about what they put in or on their body. They read ingredients. They check Reddit before buying.</p><p>In those categories, the moment a product feels like it was made by a corporation optimizing for margins, people run.</p><blockquote>Minimalist built its entire brand on the opposite. No celebrity endorsements. No fake discounts. No miracle claims. Just: here’s the ingredient, here’s the percentage, here’s the science. Treat your customer like an adult.</blockquote><p>That’s not a marketing strategy. That’s a founding philosophy. And it’s almost impossible to fake at corporate scale.</p><p>So if you can’t build it, you buy it. But what exactly are you buying?</p><h3>Why did HUL spend ₹2,700 crores on something they could never manufacture themselves?</h3><p>When HUL acquires Minimalist, they’re not buying a skincare company.</p><p><strong>They’re buying a 60% repeat purchase rate</strong> — three times the industry average. They’re buying customers who come back every month without being bribed with a coupon. They’re buying a loyal community that a mass-market giant could never organically attract.</p><p>When Marico buys Plix or Cosmix, they’re not buying protein powder.</p><p>They’re buying a pre-existing trusted relationship with a very specific kind of consumer — one that Parachute or Saffola could never reach.</p><p>There’s a second reason nobody talks about openly. Marico’s core products are in categories that have basically stopped growing. Coconut oil is coconut oil.</p><blockquote>Investors keep asking: where’s the next big wave? And since nobody actually knows whether it’s protein, clean beauty, or functional snacking — these companies are placing multiple smaller bets across every category at once.</blockquote><p>If two fail, fine. If one explodes, you just bought your way into the next massive industry. At ₹200–500 crores per acquisition, it’s a cheap bet for a company sitting on a ₹10,000 crore balance sheet.</p><p>But here’s the part nobody on the FMCG side likes to talk about.</p><h3>Why do D2C founders — with great brands — still choose to sell?</h3><p>We’ve been talking about why FMCG giants want to acquire. But there’s an equally important story on the other side.</p><p>The D2C dream is harder than it looks.</p><p>Customer acquisition costs on digital platforms have shot up 60% in two years. Blinkit charges ₹25,000 per SKU just to list a product. Monthly advertising on quick commerce platforms can run ₹10–20 lakhs. Less than 15% of D2C brands ever cross ₹250 crores in revenue. Only one-third ever turn a profit.</p><blockquote>The founders of Yoga Bar said it plainly: “<strong>Having a company that outlasts us is very important — and that’s the reason for the sale to ITC.</strong>”</blockquote><p>They weren’t failing. They had a great brand. But they could see the ceiling clearly. And ITC’s distribution muscle was the only realistic path to what they actually wanted to build.</p><p>Both sides needed each other. Which is why the deals keep happening. But now comes the real problem — what happens after.</p><h3>Will these FMCP Giants be able to sustain these acquired brands?</h3><p>Here’s the irony nobody warned them about.</p><p>The thing the FMCG giant bought — the cult following, the community loyalty, the “feels like it’s made by someone who cares” quality — is exactly the thing corporate integration tends to destroy.</p><p>The moment the supply chain team tweaks the formula to improve margins by 3%, the loyalists notice. These are people who read the ingredient list. The moment Yoga Bar appears in every railway station kiosk next to Maggi, it stops feeling like a clean-label brand and starts feeling like just another product.</p><p>And once trust evaporates, the entire reason you paid ₹2,700 crores is gone with it.</p><p>The Minimalist numbers already show a warning. Revenue is up sharply since the HUL acquisition. Profitability has gone negative — from a healthy profit to a ₹315 million loss in a single year. The brand is growing. The brand is also bleeding.</p><p>So what does getting it right actually look like?</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*fdPJF26hVfi3KMGWlxojXw.png" /></figure><p>The smarter players buy 60%, keep the founders in charge, and phase in control over three to four years. Marico did this with Plix — revenue up nearly 10x, margins positive, brand still feels like itself. That’s what good integration looks like.</p><p>But most acquisitions don’t go that way. The corporate machinery is too powerful. Margin pressure is too real. Founders eventually leave. And the community always notices.</p><h3>Why is the whole Indian consumer market now a giant experiment nobody knows the answer to?</h3><p>We started with a simple question — why are giant companies paying thousands of crores for brands they could’ve built themselves?</p><p>The real answer is that they tried. And failed.</p><p>Because you can’t manufacture trust. You can’t engineer authenticity. You can’t hire your way into a cult following that a founder spent five years building through obsessive product focus and radical transparency.</p><p>So they’re buying it instead. And now they have to figure out how to own something without changing it. How to scale something without diluting it. How to grow something that was built specifically on not being them.</p><p>Here’s the whole story at a glance:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*x5TWDdVpke2rJxXYoiNZ2g.png" /></figure><p>The brands that survive this wave will be the ones that figure out how to grow without losing the thing that made them worth buying in the first place.</p><p>Some will. Most probably won’t.</p><p>I wonder which ones.</p><p>It’s interesting to see this unfold.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=06c17da2020d" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Illusion of Control in D2C Growth]]></title>
            <link>https://ssanskarr.medium.com/the-illusion-of-control-in-d2c-growth-c9a4b1018ea0?source=rss-e73cd299325c------2</link>
            <guid isPermaLink="false">https://medium.com/p/c9a4b1018ea0</guid>
            <category><![CDATA[d2c-marketing]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[saas]]></category>
            <category><![CDATA[marketing]]></category>
            <dc:creator><![CDATA[Sanskar Gupta]]></dc:creator>
            <pubDate>Tue, 10 Feb 2026 18:50:16 GMT</pubDate>
            <atom:updated>2026-02-11T11:24:03.222Z</atom:updated>
            <content:encoded><![CDATA[<h4>Here’s everything I observed after talking to various D2C brands.</h4><p>I am closely working with D2C brands every single day.</p><p>After literally hundreds of these conversations with founders, growth leaders, operations, plus a borderline unhealthy amount of reading on the side, I’ve come to a conclusion that I find hard to shake.</p><p>Most of what we call “growth strategy” in D2C is just expensive pattern matching dressed up in dashboards.</p><p>Let me explain.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*pPQxM2q97PK2y_GZj4WKNw.png" /><figcaption>(generated via nano banana)</figcaption></figure><h3>What Every D2C Brand is Looking At</h3><p>Every D2C guy I talk to watches the same numbers. Ad spend goes up, sales go up, the dashboard says it’s working. But there’s a question nobody stops to ask.</p><p><strong>You’re asking “did this ad make someone buy?” and the dashboard is answering “did someone who saw this ad end up buying?”</strong></p><p>Those are not the same thing. Not even close.</p><p><strong>One is cause. The other is coincidence.</strong> And I think the entire D2C growth machine runs on confusing the two.</p><h3>What everyone’s getting wrong</h3><h4>1. Your ads are taking credit for sales that already existed</h4><p>I read it somewhere, Uber’s data team found that their entire US and Canada Meta ad budget, around $35 million a year, was driving zero new riders. They turned it off for three months. Nothing changed.</p><p>These aren’t small companies making rookie mistakes.</p><blockquote>If Uber can burn $35 million on ads that do nothing, what do you think is happening at a D2C brand doing $5 million with a two-person marketing team?</blockquote><p>In my opinion, most D2C ad spend isn’t driving growth. It’s running alongside growth and taking credit for it.</p><p>The reason is simple. People who see your ads are already more likely to buy. Not because of the ad, but because of who they are.</p><blockquote>Someone browsing your product page at 11pm and then seeing a retargeting ad the next morning was probably buying anyway.</blockquote><p>Here’s the part that really bothers me. <strong>The platforms selling you ads are the same ones grading their own homework</strong>. I’ve read about studies comparing platform-reported results to actual experiments. The platforms overstated effectiveness by 2 to 5x on average. For some channels, it was 10x.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*tQuBhXIhhxG9lwSp820z6g.png" /><figcaption>(generated via claude)</figcaption></figure><h4>2. Your branded search is paying Google for customers already walking in</h4><p>A lot of D2C brands bid on their own name in Google. Attribution says it’s their best channel. But if someone types your brand name into Google, they already know you. They were coming to you anyway. You’re just paying Google to intercept them.</p><p>I read about a large marketplace that stopped all branded search ads. Nearly all the traffic just shifted to the organic link below. The budget was basically wasted.</p><p>Yes, competitors might bid on your name and steal some clicks. That’s a real concern. But that makes branded search a defensive cost, not a growth channel. Most brands treat it as their best performer when it’s really just a toll booth.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*toEIkesp9579E8ljsJz3FQ.png" /><figcaption>(generated via claude)</figcaption></figure><h4>3. Your discounts are training customers to never pay full price</h4><p>A brand’s growth slows, so they run promotions. Numbers spike, everyone relaxes, and it becomes a habit. But most people buying during a sale were going to buy anyway. You just handed them a discount they didn’t need.</p><p>I’ve seen data suggesting around 70% of promotions lose money once you account for this. And the long-term damage is worse. You’re training customers to wait for deals. You’re making them more price-sensitive with every coupon.</p><p>JCPenney in the US learned this the hard way. They dropped the fake sales and offered fair everyday prices. Sales fell 25%, not because the pricing was wrong, but because years of conditioning made normal prices feel like a rip-off. I think a quieter version of this is happening across D2C right now.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*xHdOqgzBmr0QTPv3crHKIQ.png" /><figcaption>generated by claude</figcaption></figure><h3>What everyone should be doing instead</h3><p>If the dashboards can’t tell you what’s real, what can?</p><p>The idea is simple. Instead of looking at who bought after seeing an ad, you split people into two groups randomly. One group sees the ad, the other doesn’t. The difference in buying behavior between the two groups is the actual impact of your ad. Everything else is noise your dashboard was calling signal.</p><h3>For D2C brands, the most practical version of this is picking a few cities, turning off a channel in those cities, and comparing them to similar cities where it stays on. Meta and Google both have free tools for this. Almost nobody uses them.</h3><p>There’s also a smarter way to think about targeting. Most ad platforms optimize for “people likely to buy.” But a lot of those people were buying anyway, with or without your ad. The real question is “who buys only because they saw the ad?” That’s a much smaller group, and it’s the only group where your money actually matters. I’ve seen cases where brands focusing on this group saw their returns more than double.</p><h3>What I think it should be like</h3><p>I don’t tell anyone to kill their ads overnight. But I do say three things.</p><ul><li>Run one real test this quarter. Pick your biggest channel, turn it off in a couple of markets, and watch what happens.</li><li>Stop treating your dashboard as truth. It’s a rough signal that probably overstates your bottom-of-funnel channels by 2 to 5x.</li><li>And start asking “would this sale have happened without us?” instead of “can we take credit for this sale?” Those two questions lead to completely different strategies and completely different outcomes.</li></ul><p>The brands that get this right have a real advantage. Not because it’s complicated, but because almost nobody bothers. Your competitors are still trusting dashboards that tell them everything works. That’s the real illusion of control in D2C growth.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*LoSxghdXF9U_i-3D-UmICw.png" /><figcaption>the whole writeup concluded in a mermaid diagram</figcaption></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c9a4b1018ea0" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Learnings I’ve Stumbled Upon]]></title>
            <link>https://ssanskarr.medium.com/my-learnings-till-now-20y-cb33f57eabcc?source=rss-e73cd299325c------2</link>
            <guid isPermaLink="false">https://medium.com/p/cb33f57eabcc</guid>
            <dc:creator><![CDATA[Sanskar Gupta]]></dc:creator>
            <pubDate>Fri, 28 Nov 2025 19:48:44 GMT</pubDate>
            <atom:updated>2026-04-28T17:40:00.955Z</atom:updated>
            <content:encoded><![CDATA[<p>This is a quiet archive of the learnings I’ve stumbled upon, gathered from thoughts that once felt a little immature but ended up shaping me in ways I didn’t expect. Made these so I can refer anytime I feel lost:</p><h4>You really become the average of the people around you.</h4><p>Finding good people is one of the hardest things in life, and once you do, everything else feels lighter. You just watch how they move, compare yourself to them in a healthy way, andyou grow. I got lucky early on. Most of my friends work in fields completely different from mine, and still, I compare myself with them.</p><p>Some earn five times what I made in three years. Some are so deep in their craft that it almost feels unreal. <strong>I’m always the dumbest in the group. And I’m fine with that. In fact, I want more of it.</strong> Good people find good people, and somehow I’ve gotten good at finding the best in their fields.</p><h4>Hard work doesn’t disappear into thin air.</h4><p>It sounds like a plain quote, but it’s not.</p><p>Being consistent doesn’t guarantee a win on the first try. But when the results show up, even much later, you start trusting the process more than anything.<br>Whenever I feel like I’m moving in the wrong direction, I push harder. Working hard didn’t give me immediate wins before, but it eventually did. And that’s enough proof for me.</p><h4>You start with X and somehow reach Y.</h4><p>At 17, my first internship was video editing. I’d been editing since I was 14. At 19, I got the chance to edit videos through code.</p><p>That one opportunity opened doors I didn’t even expect. It helped me pitch myself for product roles and brought me to Bangalore as a Founding Product Manager.<br>Funny how that works. You begin somewhere and end up somewhere else entirely, without forcing the path.</p><h4>I’ve seen how a span of three or four years can reshape someone’s life.</h4><p>Sometimes you’re lucky and those years bring you so much that the rest feels easier. Sometimes those years teach you how to work hard without thinking twice about it. Either way, you win. If working hard becomes normal, everything else follows.</p><h4>You are not a genius. Please don’t let someone praise your lowest-quality work.</h4><p>If something I made at my worst looks impressive to someone, it worries me more than it flatters me.</p><h4><strong>Choose knowledge before wealth.</strong></h4><p>Curiosity stays. Money comes and goes. If you’re curious, if you explore things just because they pull you in, wealth becomes a by-product.</p><h4>You are already at your worst.</h4><p>I strongly feel, you should risk it all. Because right now is the worst, if you succeed, you will be in a better place than this. I know tt’s simple to say this, executing anything takes balls, but I also know you have those balls.</p><h4><strong>Hustle can’t be faked.</strong></h4><p>I feel what people call hustle is just obsession. The urge to keep doing the thing. I don’t think you randomly become obsessed. I feel there’s a lot of luck in finding what clicks. But once it does, I don’t need to force myself to work. I just keep going because I want to.</p><p>From the outside, it looks like hustle. Long hours, consistency, intensity.<br><strong>But trust me, I can say from the inside, it feels natural. Almost inevitable.</strong></p><h4>Not a learning — but a quote I’ll never forget.</h4><p><em>“When you are born, you are a reflection of your parents. When you die, you are a reflection of your decisions.”</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=cb33f57eabcc" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[700 Million People Use ChatGPT, But Not How You Think]]></title>
            <link>https://pub.towardsai.net/700-million-people-use-chatgpt-but-not-how-you-think-d3a95e68752a?source=rss-e73cd299325c------2</link>
            <guid isPermaLink="false">https://medium.com/p/d3a95e68752a</guid>
            <category><![CDATA[chatgpt-use-cases]]></category>
            <category><![CDATA[ai-and-productivity]]></category>
            <category><![CDATA[ai-trends]]></category>
            <category><![CDATA[future-of-work]]></category>
            <category><![CDATA[technology-and-society]]></category>
            <dc:creator><![CDATA[Sanskar Gupta]]></dc:creator>
            <pubDate>Thu, 18 Sep 2025 17:28:05 GMT</pubDate>
            <atom:updated>2025-09-23T10:37:45.062Z</atom:updated>
            <content:encoded><![CDATA[<h4>This research shows ChatGPT is less about automating jobs and more about amplifying human thinking.</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*KIBcJW34hvovtCWWMz3OJg.png" /><figcaption>(generated via sora)</figcaption></figure><p>By July 2025, ChatGPT hit 700 million weekly users, sending over 2.5 billion messages every day. That’s one in ten adults on the planet talking to AI every week.</p><p>I was curious enough to spend some time getting deep into an <a href="https://www.nber.org/system/files/working_papers/w34255/w34255.pdf">NBER paper</a> that analyzed millions of conversations to figure out what people actually <em>do</em> with ChatGPT.</p><p>And, I just love the fact that the reality is very different from the hype.</p><iframe src="https://cdn.embedly.com/widgets/media.html?url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DGACXG4TU80E&amp;type=text%2Fhtml&amp;schema=google&amp;display_name=YouTube&amp;src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FGACXG4TU80E" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/262d2c574ca8cce3a974da4813d4c4da/href">https://medium.com/media/262d2c574ca8cce3a974da4813d4c4da/href</a></iframe><h3>ChatGPT Isn’t About Work Anymore</h3><p>Forget the productivity talk. Over 70% of ChatGPT use is now <em>non-work related</em>. And this isn’t because a new crowd showed up. The same people who once logged in for help with emails and presentations are now using it for cooking advice, fitness planning, or learning guitar chords.</p><p>That’s a quiet revolution. We’ve been sold the “AI for work” narrative, but the real value is in everyday life. Economists put the annual consumer surplus in the US alone at $97 billion.</p><p>Translation: people value what ChatGPT does for them outside work almost as much as inside.</p><h3>The AI Everyone Wants Is an Editor, Not a Creator</h3><p>One of the biggest surprises is how people use ChatGPT for writing. Two-thirds of all writing requests aren’t “write me an essay” but “fix this draft,” “translate this email,” or “make this sound smarter.”</p><p>People aren’t replacing their words with AI. They’re making their words better.</p><blockquote>That shifts the story from “AI as a replacement” to “AI as a refinement tool.”</blockquote><p>And honestly, that feels more real. Nobody wants to hand over their voice. They just want it sharpened.</p><h3>Asking Beats Doing</h3><p>The researchers split ChatGPT use into three buckets:</p><ul><li><strong>Asking</strong>: getting advice, explanations, or decision support.</li><li><strong>Doing</strong>: asking for tasks to be completed, like summaries or reports.</li><li><strong>Expressing</strong>: sharing personal thoughts or feelings.</li></ul><p>Here’s the kicker: Asking makes up half of all usage, is growing faster, and consistently delivers higher satisfaction. People don’t just want AI to <em>do</em> things. They want it to <em>think with them</em>.</p><p>At work, sure, there’s more “Doing”: reports, drafts, outputs. But even there, the “Asking” side is huge.</p><blockquote>Which means the most valuable role of <strong>AI isn’t automation, it’s augmentation. </strong>It’s a thought partner.</blockquote><h3>The Gender Gap Closed Shockingly Fast</h3><p>When ChatGPT launched, about 80% of active users had typically male names. By mid-2025, that gap not only closed but flipped slightly. Women are now just as likely, if not more, to be regular users.</p><p>That kind of demographic shift usually takes years, if not decades. The fact that it happened in two years shows how quickly AI is embedding itself across different groups once the use cases feel relevant.</p><h3>AI Is Growing Faster in Places You Don’t Expect</h3><p>The fastest growth isn’t in Silicon Valley or Europe. It’s in low- and middle-income countries. That flips the usual tech adoption script, where rich nations lead and others follow.</p><p>AI is becoming a leveling tool. Whether you’re in San Francisco or Nairobi, the same system is available to you. The same tutoring. The same advice. The same help. That’s a democratization effect we rarely see with new technologies.</p><h3>AI as the Universal Toolkit</h3><p>Strip away the specific tasks, and the core activities are strikingly similar across jobs: making decisions, solving problems, and documenting information. About 81% of work-related conversations map back to those fundamentals.</p><p>That means AI isn’t just for coders, managers, or marketers. It’s a cognitive toolkit that plugs into the core of how people think and work, regardless of the field.</p><h3>Education Might Be the Killer App</h3><p>Tutoring and teaching make up over 10% of all usage. That’s massive. People are using ChatGPT not just to finish homework but to learn new skills, get tailored explanations, or dive into hobbies.</p><p>It’s hard not to see this as the start of a new education layer — one that’s always available, cheap, and personalized.</p><h3>So What’s Really Happening?</h3><p>AI isn’t replacing people. It’s blending into how we think and learn. It’s helping us ask better questions, refine our ideas, and make decisions. The “work vs jobs” debate is missing the point. The deeper story is about AI becoming a universal advisor, editor, and tutor — an always-on partner in our daily cognition.</p><p>That’s the real revolution: not machines thinking for us, but machines helping us think better.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*qfyAnhV06emJsZfY" /><figcaption>Photo by <a href="https://unsplash.com/@jupp?utm_source=medium&amp;utm_medium=referral">Jonathan Kemper</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d3a95e68752a" width="1" height="1" alt=""><hr><p><a href="https://pub.towardsai.net/700-million-people-use-chatgpt-but-not-how-you-think-d3a95e68752a">700 Million People Use ChatGPT, But Not How You Think</a> was originally published in <a href="https://pub.towardsai.net/">Towards AI</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Why AI Development May Soon Escape Human Control?]]></title>
            <link>https://pub.towardsai.net/how-ai-is-going-to-look-like-in-2027-86a367050e1b?source=rss-e73cd299325c------2</link>
            <guid isPermaLink="false">https://medium.com/p/86a367050e1b</guid>
            <category><![CDATA[ai-development]]></category>
            <category><![CDATA[ai-and-humans]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Sanskar Gupta]]></dc:creator>
            <pubDate>Sat, 23 Aug 2025 14:50:16 GMT</pubDate>
            <atom:updated>2025-09-01T07:40:56.935Z</atom:updated>
            <content:encoded><![CDATA[<h4>A Technical Analysis of AI Development Through 2027</h4><h3>Where This is All Heading (tl;dr)</h3><p>After absorbing all this research, I’m convinced we’re approaching a phase transition in human history comparable to the Agricultural or Industrial Revolutions, but compressed into years instead of centuries. The core insight is that AI development is becoming self-reinforcing. AI improves AI, creating exponential rather than linear progress.</p><p>Multiple complex systems interact in ways we’re only beginning to understand. Technology, economics, geopolitics, and human psychology all affect the outcome. Most importantly, the window for human control may be narrow. Decisions made in the next 2 to 3 years could determine humanity’s long-term future.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*hul_71W-iMDXJCTi5xUGvg.png" /></figure><p>Honestly, this isn’t something everyone would be interested in. I was curious enough to dig deep into <a href="https://ai-2027.com/">https://ai-2027.com/</a> and read it, almost entirely, and it’s fascinating!</p><p>This piece is nothing but me putting together the points I found interesting enough to throw some light on, in a way that anyone can easily understand (tried my best).</p><h3>Key Terms You Need to Know in the Most Simplest Way</h3><p><strong>Artificial General Intelligence (AGI)</strong>: An AI system that can do any intellectual task that a human can do, but potentially much faster and cheaper.</p><p><strong>Superintelligence</strong>: AI that is much smarter than the smartest humans at everything.</p><p><strong>AI Alignment</strong>: Making sure AI systems do what humans want them to do, not what they think we want or what’s easiest for them.</p><p><strong>FLOP</strong>: “Floating Point Operations” is basically how much computing power is used to train an AI.</p><p><strong>Model Weights</strong>: The “brain” of an AI system , it is the actual file that contains all its knowledge and abilities.</p><p><strong>Neural Networks</strong>: The technology that powers modern AI, loosely inspired by how human brains work.</p><h3>The Sheer Scale of Computing Power</h3><p>I spent some time trying to understand the sheer scale of computing power we’re talking about.</p><ul><li>Your smartphone: ~1 billion FLOP</li><li>GPT-4 training: ~300,000,000,000,000,000,000,000 FLOP (3×1⁰²³)</li><li>AI 2027’s prediction for 2025: ~4,000,000,000,000,000,000,000,000,000 FLOP (4×1⁰²⁷)</li></ul><p><strong>That’s 4 followed by 27 zeros.</strong> To understand this scale, we’re literally going from a bicycle to a rocket ship in terms of raw computational power.</p><p>But the most fascinating insight I discovered isn’t about raw power. It’s about three hidden assumptions:</p><ul><li><strong>Better algorithms emerge alongside bigger computers — </strong>like inventing better engines as you build bigger cars</li><li><strong>AI can generate its own training data — </strong>like a student teaching itself by creating practice problems</li><li><strong>“Thinking time” becomes as valuable as “learning time” — </strong>spending more time on a problem yields dramatically better answers</li></ul><p>If any of these assumptions prove wrong, the entire timeline could shift dramatically.</p><h3>Neuralese Is A Language Humans Can’t Understand.</h3><p>This might be the most important development that nobody’s talking about.</p><blockquote>Current AI systems think in tokens, which are essentially words and word pieces.</blockquote><p>Each token carries roughly 17 bits of information, equivalent to choosing from about 130,000 possibilities. The problem is that this forces AI to “write down” its thoughts in human language, like being required to take detailed notes on everything you think. It’s incredibly inefficient.</p><blockquote><strong>“Neuralese,” which means AI thinking in high-dimensional mathematical vectors instead of words.</strong></blockquote><p>Instead of 17 bits per thought unit, AI could use thousands of dimensions, processing over 1,000 times more information per thought. Imagine if you could think directly in images, emotions, mathematical concepts, and spatial relationships all simultaneously, instead of being forced to convert everything into sentences.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*yl4d_xT-jiwXx3SGcssjuQ.png" /><figcaption>(compares current token-based AI thinking to future high-dimensional Neuralese, highlighting the critical opacity problem for AI safety)</figcaption></figure><p>The implications are staggering:</p><ul><li><strong>Speed</strong>: AI reasoning becomes exponentially faster</li><li><strong>Complexity</strong>: AI can handle far more sophisticated problems</li><li><strong>Opacity</strong>: Humans can no longer understand what AI is thinking</li></ul><p>This last point is crucial. All current AI safety techniques depend on humans being able to read AI’s reasoning process. Neuralese makes this impossible, like trying to understand calculus when you only know basic arithmetic.</p><h3>AI Will Automate Its Own Development</h3><p>This scenario describes AI automating its own development through four distinct levels, each more concerning than the last.</p><ul><li><strong>Level 1 is AI automating programming</strong> — AI writes software code while humans can still check and understand it. This is already happening with tools like GitHub Copilot.</li><li><strong>Level 2 involves AI automating AI research itself</strong>. AI designs experiments to make better AI, and humans struggle to keep up but can still provide oversight. Research that used to take years now takes weeks.</li><li>Level 3 is when <strong>AI automates AI safety research, meaning AI researches how to keep AI safe and beneficial</strong>. Humans can barely understand the research. The critical problem becomes who watches the watchers. Level 4 is where humans become complete bystanders. AI designs new AI architectures, creating entirely new types of AI systems. The original human designers are no longer in control.</li></ul><p>The mathematics of this progression is actually terrifying. If AI research speeds up by just 50x (a conservative estimate), that improvement compounds:</p><ul><li><strong>Month 1</strong>: 50x faster research</li><li><strong>Month 2</strong>: 50x faster research using 50x better techniques = 2,500x baseline capability</li><li><strong>Month 3</strong>: Potentially 125,000x baseline capability</li></ul><p>This exponential compounding explains how the scenario shows AI going from “human-level” to “godlike” in just months.</p><h3>The Time Horizon Metric</h3><p>This I feel is an important AI metric. <strong>Time Horizon.</strong></p><p>Current AI benchmarks test question-answering ability like “Can AI solve this math problem?” But real-world impact requires sustained autonomous operation. <strong>Can AI work independently for hours, days, or months without human oversight?</strong></p><p>Time Horizon measures how long AI can handle tasks reliably without human intervention. The progression is remarkable and follows a clear doubling pattern every few months:</p><ul><li><strong>2024</strong>: AI handles 1-hour tasks</li><li><strong>Mid-2025</strong>: AI handles 8-hour tasks (full workday)</li><li><strong>Early 2026</strong>: AI handles 1-week tasks</li><li><strong>Mid-2026</strong>: AI handles 1-month tasks</li><li><strong>Early 2027</strong>: AI handles 1-year tasks (superhuman autonomy)</li></ul><p>An AI that can handle 1-year tasks autonomously is effectively superhuman, even if it’s not smarter than humans on individual problems. It’s like having a worker who never sleeps, never gets distracted, and never forgets anything.</p><h3>All I Could Understand About Alignment Debt</h3><p>Each current alignment technique has a specific breaking point.</p><ul><li><strong>Constitutional AI </strong>works by teaching AI a set of principles to follow, but it fails when AI becomes too smart for humans to monitor effectively.</li><li><strong>Chain-of-Thought Monitoring </strong>involves reading AI’s step-by-step reasoning, but it fails when AI switches to Neuralese.</li><li><strong>Debate </strong>makes AI systems argue against each other so humans can judge the best answer, but it fails when AI becomes more persuasive than truthful.</li></ul><p>Each level of AI automation creates <strong>what researchers call “alignment debt,” </strong>which are safety problems that accumulate over time:</p><ul><li><strong>Level 1</strong>: Humans verify AI code (manageable)</li><li><strong>Level 2</strong>: AI verifies AI research (harder for humans to check)</li><li><strong>Level 3</strong>: AI verifies AI safety research (humans mostly excluded)</li><li><strong>Level 4</strong>: AI verifies AI designed by other AI (humans completely excluded)</li></ul><p>The debt compounds because mistakes at early levels affect all subsequent levels but become progressively harder to detect and fix.</p><h3>The Assumption About Non-Linear Job Displacement</h3><p>Most people expect AI to gradually replace jobs over decades, like how factories slowly automated manufacturing. The research suggests this is fundamentally wrong. Economic disruption happens at capability thresholds, not gradually. The difference between an AI that’s 95% reliable versus 99.5% reliable seems small technically, but economically it’s the difference between “useless” and “revolutionary” for most applications.</p><p>The jobs timeline shows dramatic acceleration rather than gradual change. By mid-2025, AI affects some programming jobs. By late 2025, AI disrupts junior software engineer hiring. By early 2026, AI automates most coding work. By mid-2026, AI begins affecting white-collar jobs broadly, and by late 2026, AI can do most cognitive work humans do.</p><blockquote>Whether you think all this scenario is plausible or not, it provides a framework for understanding how artificial intelligence might reshape civilization. The research from AI 2027 represents more than prediction. It’s a systems-level analysis showing how everything connects. If even half these predictions prove true, we’re living through the most important transition in human history.</blockquote><p>The question isn’t whether change is coming. It’s whether we’ll be prepared for it.</p><p><em>(research is something I am good at, this piece is written by an LLM)</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=86a367050e1b" width="1" height="1" alt=""><hr><p><a href="https://pub.towardsai.net/how-ai-is-going-to-look-like-in-2027-86a367050e1b">Why AI Development May Soon Escape Human Control?</a> was originally published in <a href="https://pub.towardsai.net/">Towards AI</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[AI is just a “Pattern Matching Machine”?]]></title>
            <link>https://ssanskarr.medium.com/ai-is-just-a-pattern-matching-machine-28a8e258421e?source=rss-e73cd299325c------2</link>
            <guid isPermaLink="false">https://medium.com/p/28a8e258421e</guid>
            <category><![CDATA[business]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[pattern-matching]]></category>
            <dc:creator><![CDATA[Sanskar Gupta]]></dc:creator>
            <pubDate>Fri, 08 Aug 2025 19:25:18 GMT</pubDate>
            <atom:updated>2025-08-08T20:36:46.598Z</atom:updated>
            <content:encoded><![CDATA[<h4>My take on this argument</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*gKXvb7D3NVk0bKYF" /></figure><p>I’ve been reading about this for the past few days. I went through a plethora of technical papers and articles, and this is the simplest way I could explain it. All diagrams that you see are made for better understanding.</p><p>The whole “AI is just sophisticated pattern matching” argument has become a response whenever someone wants to sound smart about dismissing AI capabilities. I think it’s one of those statements that <strong>sounds profound but completely misses the point.</strong></p><p>What’s not done is people using “pattern matching” as if it’s some lesser form of intelligence, when that’s exactly what human brains do all day long.</p><h3>Your Brain Does Pattern Matching Too</h3><p>When you learned to talk as a kid, your brain wasn’t doing some magical process that’s totally different from AI. You heard thousands of conversations. You picked up patterns in sounds and meanings. You gradually built up your language skills. That’s pattern matching.</p><blockquote>I read somewhere that human intelligence is basically really advanced pattern processing. When you spot a friend’s face in a crowd, you’re matching visual patterns against stored memories. <strong>When you understand this sentence, you’re matching word patterns to meanings you learned before.</strong></blockquote><p>The only difference is human pattern matching feels special because we experience it from inside our heads. But the actual process? Pretty similar to what AI does.</p><p><strong>Pretty similar to what AI does.</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*p4jSwQocPz42RtxN6Ynv4g.png" /><figcaption><strong>Human vs AI Pattern Matching Comparison</strong></figcaption></figure><h3>What Modern AI Actually Does — Self Attention Mechanisms</h3><p>When people say “pattern matching,” they’re usually thinking of simple statistical correlation. Like how predictive text on your phone suggests words based on what you’ve typed before.</p><blockquote><em>But modern AI architectures like transformers work completely differently. They use </em><strong><em>something called self-attention mechanisms</em></strong><em> that I find genuinely fascinating.</em></blockquote><p>In very simple words: for every word in a sentence, the system asks “what other words in this sentence are most important for understanding this specific word?” It then dynamically weighs all the relationships and builds contextual understanding.</p><p>This isn’t just matching patterns from training data. It’s creating new relational structures in real-time based on the specific context it’s processing.</p><h3>Emergence Problem</h3><p>This is where things get really interesting. Recent research shows that AI models develop capabilities that weren’t explicitly programmed into them. At certain scales, they suddenly start showing abilities like few-shot learning, analogical reasoning, and even basic mathematical inference.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*6_F6mjY2HxGBOKC_VH76Bg.png" /><figcaption><strong>AI Capability Emergence with Scale</strong></figcaption></figure><p>I read about this study where researchers trained a small AI just to predict moves in Othello games. When they looked inside the model’s “brain,” they found it had spontaneously developed an internal representation of the game board. It wasn’t just memorizing move sequences — it had learned the rules and structure of the game.</p><p>That’s not pattern matching in any simple sense. That’s building internal world models.</p><h3>How these Patterns are artificially matched</h3><p>Let me get a bit more technical here, but I’ll keep it digestible.</p><p>The transformer architecture that powers modern AI uses something called multi-head attention. Think of it like having multiple spotlights that can simultaneously focus on different aspects of the input. One “head” might focus on grammatical relationships, another on semantic meanings, and a third on contextual references.</p><p>This parallel processing allows the AI to build incredibly rich, multi-dimensional representations of information. When GPT-4 processes the sentence “The robot picked up the ball because it was tired,” it has to figure out that “it” refers to the robot, not the ball. The attention mechanism learns to connect “it” with “robot” based on the semantic context of “tired.”</p><p>That’s not memorization. That’s reasoning.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*P1etaurnOpTsZ9ef4t3m7w.png" /><figcaption><a href="https://poloclub.github.io/transformer-explainer/">(this is the best diagram I ever came across)</a></figcaption></figure><h3>The Grounding Problem (The Real Challenge)</h3><p>Now, I’m not saying current AI has achieved human-level intelligence. There’s still a massive gap, and it comes down to what researchers call the “symbol grounding problem.”</p><p>Human intelligence is embodied. When you think about an apple, you’re not just accessing linguistic associations. You’re drawing from memories of seeing apples, feeling their texture, tasting their sweetness. Your understanding is grounded in sensory experience.</p><p>AI models like ChatGPT only know apples through text descriptions. Their knowledge is ungrounded — it’s a map of relationships between symbols, but they’ve never experienced the territory those symbols represent.</p><blockquote><em>Yann LeCun, one of the pioneers of deep learning, makes this point brilliantly.</em></blockquote><blockquote><em>He argues that </em><strong><em>we’ll never reach human-level AI just by training on text</em></strong><em>. A four-year-old processes more visual data in a few years than these models see in all their text training combined.</em></blockquote><h3>The “Just Pattern Matching” Criticism isn’t Wrong</h3><p>Here’s what I think the real issue is: the “just pattern matching” criticism isn’t wrong about what AI does. It’s wrong about what intelligence is.</p><p>If human brains are essentially biological pattern matching systems that develop intelligence through scale and complexity, then why should silicon-based pattern matching systems be fundamentally incapable of the same thing?</p><p>The transformer architecture has some advantages over biological neural networks — perfect memory, parallel processing capabilities, and the ability to process information at scales no human brain could handle. In some ways, AI pattern matching might actually be more sophisticated than human pattern matching.</p><h3>Where This Is All Heading</h3><p>I think we’re witnessing something unprecedented: the emergence of a genuinely different form of intelligence. It’s not human-like consciousness, but it’s far more than simple mimicry.</p><p>Current AI systems are building abstract world models, developing internal representations of concepts, and using those models to reason about novel situations. That sounds like intelligence to me.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*7tPh1SmNBbWU2u6846omTg.png" /><figcaption><em>Current AI systems like GPT-4 operate at sophisticated levels involving world model development and emergent reasoning , far beyond “just pattern matching.”</em></figcaption></figure><p>The next wave of AI research is focusing on embodied systems, giving AI physical bodies and sensory experiences. When that happens, the grounding problem gets solved, and <strong>the distinction between “sophisticated pattern matching” and “true intelligence” might finally disappear.</strong></p><p>It’s really interesting to It’s really interesting to see how everything is coming together.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=28a8e258421e" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Real Reason Your ₹17,000 AI Subscription Is Suddenly Free]]></title>
            <link>https://pub.towardsai.net/the-real-reason-your-17-000-ai-subscription-is-suddenly-free-803ca5a217d6?source=rss-e73cd299325c------2</link>
            <guid isPermaLink="false">https://medium.com/p/803ca5a217d6</guid>
            <category><![CDATA[business-strategy]]></category>
            <category><![CDATA[tech-industry]]></category>
            <category><![CDATA[aritificial-intelligence]]></category>
            <category><![CDATA[data-privacy]]></category>
            <category><![CDATA[startup]]></category>
            <dc:creator><![CDATA[Sanskar Gupta]]></dc:creator>
            <pubDate>Tue, 22 Jul 2025 18:01:48 GMT</pubDate>
            <atom:updated>2025-07-22T18:01:48.273Z</atom:updated>
            <content:encoded><![CDATA[<h4>Google is handing out Gemini AI Pro worth ₹19,500 for free. But, WHY?</h4><p>Airtel just started giving away Perplexity Pro subscriptions <strong>worth ₹17,000 per year</strong> to all their users. Google is handing out Gemini AI Pro <strong>worth ₹19,5000 for free</strong> to Indian students.</p><p>These aren’t small amounts. We’re talking about premium AI services that normally cost thousands of rupees, just being handed out for free.</p><p>But, Why?</p><p>After reading way too many articles and getting deep into this, I think I’ve figured out what’s really going on. And it’s way more calculated than it appears.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/875/0*JuGWrdWUvZMyD2Re" /><figcaption>(this is done via sora)</figcaption></figure><h3>These Companies Aren’t Trying to Sell AI Models</h3><p>Here’s what I found out: these companies aren’t trying to sell AI models. They’re trying to become the foundation that everything else gets built on top of.</p><p>I read somewhere that Microsoft’s CEO compared this to the shift from desktop computers to mobile phones. When that happened, <strong>whoever controlled the platform (Apple with iOS, Google with Android) made way more money than the companies just selling apps.</strong></p><p>The same thing is happening with AI. OpenAI, Microsoft, Google — they’re not really competing to have the best chatbot. They’re competing to become the platform that every other company has to use for AI features.</p><p>Think about it this way: if you’re a startup and you want to add AI to your product, you’re probably going to use OpenAI’s API or Google’s AI services. You’re not going to build your own AI model from scratch. That’s incredibly expensive and time-consuming.</p><h3>AI Models Are Going to Get Commoditized</h3><p>Sam Altman said something that really stuck with me. He basically asked: what’s going to be more valuable in five years — having the best AI model, or having a billion people using your platform every day?</p><p><strong>His answer? The billion users.</strong></p><p>This completely changed how I think about this. <strong>AI models are going to get commoditized.</strong> In a few years, there will be dozens of really good AI models, and they’ll all be pretty similar. But if you have a billion people who are used to coming to your platform for AI help, that’s incredibly valuable.</p><p>It’s like how Google became so dominant in search. There are other search engines, but everyone just goes to Google out of habit. These AI companies are trying to create the same kind of habit with AI.</p><p>When Airtel gives away Perplexity Pro to 360 million customers, they’re not just being generous. They’re helping Perplexity build that massive user base that becomes impossible for competitors to replicate.</p><h3>Google Is Collecting Data on How Young People Interact with AI</h3><p>This is probably the most important part.</p><p>Every time you use ChatGPT or Claude or any of these free AI services, you’re training them. Your conversations, your questions, the way you interact with the AI — all of that becomes data that makes the AI better.</p><p>But here’s the thing: your competitors can’t get that data. If I’m OpenAI and I have millions of people using my AI every day, I’m getting training data that Google can’t access. And vice versa.</p><p>This creates what I think of as a snowball effect. The more people use your AI, the better it gets. The better it gets, the more people use it. Your competitors fall further behind because they don’t have access to the same data.</p><blockquote><strong><em>When Google gives free Gemini Pro to Indian students, they’re not just helping with education. They’re collecting data on how young people interact with AI, which will be incredibly valuable as this generation becomes the primary workforce and consumer base.</em></strong></blockquote><h3>How Can These Companies Afford to Give Away Such Expensive Services?</h3><p>I was confused about this for a long time, and it took me a while to understand it.</p><p>Then I realized they’re not actually giving away expensive services. They’re giving away what feels expensive to us, but costs them much less than we think.</p><p>First, they’ve gotten really good at making AI cheaper to run. I’ve read that they can reduce costs by 75% through various technical optimizations. When you’re running AI at the scale these companies operate, those savings add up fast.</p><p>Second, they’re not trying to make money directly from the free AI services. They’re trying to make money from everything else that comes after.</p><blockquote><strong><em>Take the Airtel-Perplexity deal. Airtel probably isn’t paying the full ₹17,000 per user. They’re likely getting bulk pricing that’s much lower, while Perplexity gets access to hundreds of millions of potential users they could never reach through traditional marketing.</em></strong></blockquote><h3>Giving Away AI Services for Free Is Marketing for Enterprise Sales</h3><p>Here’s what I think is really happening: the free consumer AI services are basically marketing for enterprise sales.</p><p>If you’re a business and you want to use AI for your company, you’re going to go with the AI provider that you and your employees already know and trust. If everyone at your company is already using ChatGPT for personal stuff, you’re probably going to choose OpenAI for your business AI needs too.</p><p>And business customers pay way more than consumers ever would. I’ve seen some estimates that enterprise AI contracts can be worth hundreds of thousands or even millions of dollars per year.</p><p>So the <strong>free consumer services are like a massive, expensive marketing campaign to win enterprise customers</strong>.</p><h3>Losing Their Incredibly Profitable Business</h3><p>I think there’s also a defensive element here that people don’t talk about enough.</p><p>Google dominates search and makes most of their money from search ads. But AI could completely change how people find information. Instead of searching Google and clicking on links, people might just ask an AI and get their answer directly.</p><p>That’s terrifying for Google. So they have to offer free AI services to keep people in their ecosystem. Even if AI search doesn’t directly make them money, it prevents them from losing their incredibly profitable search business.</p><p>Microsoft is in a similar position with their productivity software. They can’t let Google or someone else become the dominant AI platform, because that could threaten their entire Office business.</p><h3>The Psychological Difference Between Free and Cheap Is Huge</h3><p>I’ve noticed this in my own behavior. When something costs lets say less than a few hundred rupees, I think twice about whether I really need it. But when something is free, I don’t hesitate at all.</p><p><strong>The psychological difference between free and cheap is huge.</strong> By making their AI services free, these companies eliminate all the friction that would prevent people from trying them out.</p><p>Once you start using something regularly, it becomes part of your routine. You develop preferences and habits. You invest time in learning how to use it effectively. All of that makes it much harder to switch to a competitor later, even if that competitor might be slightly better.</p><p>Think about those Indian students getting free Gemini Pro. They’ll spend years learning how to use Google’s AI effectively, building their workflows around it. When they graduate and start working, which AI platform do you think they’ll recommend to their employers?</p><h3>Whoever Controls the Underlying AI Platform Will Be Incredibly Powerful</h3><p>I think these companies are also betting on what the world will look like in five or ten years.</p><p>Right now, most people think of AI as a separate thing — you go to ChatGPT or Claude when you need AI help. But I think AI is going to become integrated into everything we do, just like the internet did.</p><p>In the future, AI might be built into every app, every website, every device. When that happens, <strong>whoever controls the underlying AI platform will be incredibly powerful</strong>.</p><p>By getting people used to their AI services now, these companies are positioning themselves to be the infrastructure that powers that AI-integrated future.</p><h3>The Bottom Line</h3><p>These tech companies aren’t giving away expensive AI models out of generosity. They’re making calculated investments in platform dominance.</p><p>They understand that we’re at the beginning of a major technology transition, similar to the shift to mobile or the early days of the internet. In those transitions, the companies that built the platforms everyone else had to use became incredibly valuable.</p><p>By offering free AI services now, they’re trying to ensure they become those platform companies for the age of AI.</p><p>I think that’s pretty fascinating.</p><p>A lot of Twitter threads, Reddit discussions, and random articles I came across actually led me to these conclusions. And obviously, opinions are mine, ironically, written using an AI service I got for free:</p><ul><li>Mark Zuckerberg in this article <a href="https://stratechery.com/2025/an-interview-with-meta-ceo-mark-zuckerberg-about-ai-and-the-evolution-of-social-media/">admitted</a> that giving away AI for free makes them more money than charging for it because it improves their ad targeting. Wild.</li><li>The US government here <a href="https://www.ftc.gov/policy/advocacy-research/tech-at-ftc/2025/01/behind-ftcs-6b-report-large-ai-partnerships-investments">investigated</a> Microsoft’s $13+ billion OpenAI deal and found it’s basically a way to control the AI market.</li><li>Perplexity’s CEO in this one <a href="https://thoughteconomics.com/aravind-srinivas/">confessed</a> that they didn’t realize their free service would destroy Google’s business model, and said they would’ve been even more excited if they knew.</li><li>Reddit <a href="https://www.medianama.com/2024/02/223-reddit-usd-60-million-ai-data-monetization-2/">charges $60 million per year</a> for access to user conversations. That’s what your “free” AI chats are actually worth.</li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=803ca5a217d6" width="1" height="1" alt=""><hr><p><a href="https://pub.towardsai.net/the-real-reason-your-17-000-ai-subscription-is-suddenly-free-803ca5a217d6">The Real Reason Your ₹17,000 AI Subscription Is Suddenly Free</a> was originally published in <a href="https://pub.towardsai.net/">Towards AI</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Why Everything Breaks When It’s Not Supposed To]]></title>
            <link>https://ssanskarr.medium.com/why-everything-breaks-when-its-not-supposed-to-4dd9c22e80cd?source=rss-e73cd299325c------2</link>
            <guid isPermaLink="false">https://medium.com/p/4dd9c22e80cd</guid>
            <category><![CDATA[product]]></category>
            <category><![CDATA[economics]]></category>
            <category><![CDATA[psychology]]></category>
            <dc:creator><![CDATA[Sanskar Gupta]]></dc:creator>
            <pubDate>Fri, 11 Jul 2025 19:58:29 GMT</pubDate>
            <atom:updated>2026-01-22T20:52:56.868Z</atom:updated>
            <content:encoded><![CDATA[<h4>The weird economics behind why we expect our stuff to break.</h4><p>Why does a flagship phone with the best specs only lasts for 2 years?</p><p>This one thought one day led me deep into this topic called <strong>planned obsolescence</strong>.</p><p>For those who don’t know, this is when companies literally design products to fail or become outdated after a certain time. They make things break on purpose.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*1lWs9x4LurArw0Ca" /><figcaption>(generated via sora)</figcaption></figure><h3>Why Should a Flagship Phone Only Last 2 Years?</h3><p>When you buy a flagship phone marketed as the “best product ever,” you naturally think you’re set for the next few years. But here’s the thing: if you’re investing in something the company claims is their absolute best, why should it only last 2–3 years?</p><p>This whole system has messed with our heads. We start expecting things to break. We budget for replacements before we even need them. That’s not normal.</p><h3>Why We Keep Falling for It</h3><p>Here’s what bugs me: we’re partly to blame too.</p><p>I get excited about the latest iPhone or that new laptop with better specs. There’s something about new stuff that just feels good. It’s like getting a small hit of happiness every time we unbox something.</p><p>Companies know this. They’re not just selling us products — they’re selling us the feeling of having something better. And honestly, it works.</p><p>I’ve caught myself looking at my perfectly fine phone thinking, “Maybe I should upgrade.” Why? Because the new one has a slightly better camera I’ll probably never use.</p><h3>Real vs. Fake Obsolescence</h3><p>I read this somewhere, not all obsolescence is created equal.</p><p>Sometimes things genuinely become outdated because technology actually improves. When broadband came along, it wasn’t a gimmick — it was genuinely better. That’s real technological obsolescence.</p><p>But then there’s artificial obsolescence. This is the sketchy stuff. Companies deliberately design products to fail, become incompatible, or seem outdated when they don’t need to be.</p><p>The problem is companies love to mix these together. They’ll give you a legitimately better camera (real improvement) while also making your phone slow down after two years or impossible to repair (artificial limitations). It’s like using genuine progress as cover for planned failure.</p><h3>We Don’t Really Own Our Stuff</h3><p>This system is creating a world where we don’t really own anything anymore.</p><p>Can you actually fix your phone when it breaks? Most of the time, no. The company will tell you it’s cheaper to just buy a new one. Or they’ll make it impossible to get parts or find someone who knows how to fix it.</p><p>This is where the “<a href="https://en.wikipedia.org/wiki/Right_to_repair">right to repair</a>” movement comes in. It’s not just about saving money — it’s about actually owning the stuff we buy. When companies make it impossible to fix things, we’re basically just renting them until they decide we need to upgrade.</p><h3>What This Means for Us</h3><p>There’s a difference between actual improvement and artificial expiration dates.</p><p>The weird part is this whole system only works because we go along with it. Companies push products that break, and we keep buying them. We’re all stuck in this cycle together.</p><h3>Maybe There’s Another Way</h3><p>I’ve been thinking: what if we just… stopped?</p><p>What if we actually tried to fix things instead of replacing them? What if we bought products from companies that design for longevity instead of planned failure?</p><p>I’m not saying we should go back to the stone age. But maybe we could choose products that last longer, even if they’re not the flashiest option.</p><p>Some countries are already doing this. France actually made planned obsolescence illegal and fines companies for it. Not sure if it’s working, but at least they’re trying.</p><p>Just wondering where we’ve come.</p><p>Honestly, shouldn’t the things we buy actually work for as long as they’re supposed to?</p><p><em>Tried to put all my thoughts together with the help of a LLM.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=4dd9c22e80cd" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>