<?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 doanythingnow on Medium]]></title>
        <description><![CDATA[Stories by doanythingnow on Medium]]></description>
        <link>https://medium.com/@doanything.now?source=rss-6e21fed3831f------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*f8nyovtfj6byRMih</url>
            <title>Stories by doanythingnow on Medium</title>
            <link>https://medium.com/@doanything.now?source=rss-6e21fed3831f------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Fri, 20 Nov 2026 00:00:09 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@doanything.now/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[Rethinking the Cloud: The Sustainability and Economic Power of Web3 Storage]]></title>
            <link>https://medium.com/@doanything.now/rethinking-the-cloud-the-sustainability-and-economic-power-of-web3-storage-06cf1a81ebc5?source=rss-6e21fed3831f------2</link>
            <guid isPermaLink="false">https://medium.com/p/06cf1a81ebc5</guid>
            <category><![CDATA[clean-energy]]></category>
            <category><![CDATA[green-tech]]></category>
            <category><![CDATA[green-energy]]></category>
            <category><![CDATA[decentralization]]></category>
            <category><![CDATA[decentralized-finance]]></category>
            <dc:creator><![CDATA[doanythingnow]]></dc:creator>
            <pubDate>Sat, 15 Aug 2026 19:58:01 GMT</pubDate>
            <atom:updated>2026-08-15T19:59:09.795Z</atom:updated>
            <content:encoded><![CDATA[<h4>In Web3, users own their data, identities, and digital assets. Decentralized Storage ensures that this data remains secure, private, and accessible without relying on centralized companies</h4><p>Every time you back up a database, stream media, or upload a file to the cloud, a server farm hums — and burns electricity. While digital transformation is often framed as “paperless” and inherently green, legacy cloud giants like AWS, Azure, and Google Cloud are quietly consuming vast amounts of global power while driving up enterprise infrastructure budgets. But for real:</p><p>What if the future of data storage wasn’t building bigger, hotter server farms, but tapping into a smart, decentralized network?</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*dLG1ibIPOaUmskXZ" /><figcaption>Photo by <a href="https://unsplash.com/@tarekbadrph?utm_source=medium&amp;utm_medium=referral">Tarek Badr</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>Moving to Web3 storage isn’t just an architectural shift — it is a financial and environmental imperative. By converting idle, existing hard drive space into a secure, distributed cloud, Web3 protocols are solving two of enterprise tech’s biggest headaches at once: astronomical cloud bills and massive carbon footprints.</p><p>Traditional cloud providers like AWS, Azure, and Google Cloud rely on massive server farms that consume roughly 1–3% of all global electricity. Expanding these centralized data centers requires continuous land development, heavy cooling systems, and massive supply chain emission. Manufacturing a single 1TB hard drive alone generates nearly 20 kg of carbon!</p><p>Decentralized storage flips this model by turning globally distributed, underutilized hard drive capacity into a unified, secure cloud network.</p><h4><strong>Environmental Benefits</strong></h4><p>Instead of building and powering new server farms, decentralized Web3 networks tap into existing idle hardware that is already powered and running in servers worldwide.</p><ul><li><strong>Slashing Carbon Impact:</strong> Research from <a href="https://medium.com/u/31bb19d6e641">Storj Labs</a> shows that distributed cloud storage can reduce carbon emissions by up to 83% over three years compared to traditional centralized providers by eliminating redundant hardware manufacturing and unnecessary multi-region data cloning.</li><li><strong>Verifiable Clean Energy:</strong> Unlike energy-wasteful Proof-of-Work blockchains, networks like Filecoin use Proof-of-Useful-Work algorithms (such as Proof-of-Spacetime) where energy directly yields usable storage capacity. Initiatives like Filecoin Green track renewable energy inputs on-chain, enabling organizations to audit and prove their zero-carbon data footprint.</li><li><strong>Curtailing E-Waste:</strong> Reusing existing enterprise hard drives extends hardware lifespans, directly reducing electronic waste and keeping functional equipment out of landfills.</li></ul><h4><strong>Economic Impact: Demolishing Costs &amp; Monetizing Idle Capacity</strong></h4><p>Decentralized storage transforms data management from a centralized monopoly into a hyper-efficient, open marketplace.</p><ul><li><strong>70–80% Lower Storage Costs:</strong> Traditional enterprise cloud storage often costs around $23 per terabyte per month. Decentralized storage marketplaces lower prices to approximately $2–$5 per terabyte per month through open competition.</li><li><strong>Monetizing Stranded Infrastructure:</strong> Data centers, businesses, and node operators can rent out unused storage capacity, converting underutilized hardware into active income streams.</li><li><strong>Eliminating Predatory Egress Fees:</strong> Legacy cloud providers frequently lock organizations in with steep fees when retrieving or transferring data out. Web3 storage protocols eliminate these lock-in mechanics with transparent, auditable billing verified directly on-chain. <a href="https://medium.com/u/b4f94395474f">Serverion</a> provides hosting services, including web hosting, VPS, dedicated servers, and colocation. Check <em>them</em> out.</li></ul><h4><strong>The Future of Storage is Green and Distributed</strong></h4><p>The migration to decentralized storage is no longer a theoretical promise — it is happening right now. Whether you are an enterprise IT leader looking to slash cloud overhead, a developer building resilient dApps, or a sustainability officer striving to hit ambitious ESG goals, Web3 storage delivers a rare win-win: lower costs and lower carbon emissions.</p><p>Stop letting legacy cloud monopolies lock in your data and lock out sustainability. Audit your organization’s current cloud expenditure today, run a pilot on decentralized networks like Filecoin or Storj, and take the first step toward a leaner, greener digital infrastructure.</p><p>I’m <a href="https://medium.com/u/6e21fed3831f?source=post_page---user_mention--43a4b0afdf93---------------------------------------">Dan Ikigai</a>, you can <a href="https://x.com/ikigai_dan">follow me</a> on X, I also run a <a href="https://www.publish0x.com/doanythingnow">personal blog</a>, I host an IPFS <a href="https://bafybeida4re5vekb7gzu4y5jlgi367bxxixzkjgem6ruz5l5cf2xgnnflq.ipfs.dweb.link/?filename=Spider-Man_Brand_New_Day_2026.mp4">node</a> that I keep loaded with useful goodies for the weary wanderer, and I look forward to building a community to interact with!</p><p>:)</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=06cf1a81ebc5" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[THORChain Exploit Report: The Three-Part Attack]]></title>
            <link>https://medium.com/@doanything.now/thorchain-exploit-report-the-three-part-attack-43a4b0afdf93?source=rss-6e21fed3831f------2</link>
            <guid isPermaLink="false">https://medium.com/p/43a4b0afdf93</guid>
            <category><![CDATA[hacking]]></category>
            <category><![CDATA[cryptocurrency-news]]></category>
            <category><![CDATA[exploit]]></category>
            <category><![CDATA[defi]]></category>
            <category><![CDATA[blockchain-technology]]></category>
            <dc:creator><![CDATA[doanythingnow]]></dc:creator>
            <pubDate>Sat, 15 Aug 2026 14:31:48 GMT</pubDate>
            <atom:updated>2026-08-15T14:31:48.296Z</atom:updated>
            <content:encoded><![CDATA[<h4>Cryptographic flaw in THORChain’s TSS implementation allowed three vulnerabilities to grant hackers $10.8 million</h4><p>On <a href="https://coinpedia.org/crypto-live-news/thorchain-exploit-drains-7-4m-across-bitcoin-ethereum-bsc-and-base/amp/">May 15, 2026, THORChain suffered a $10.8 million exploit across multiple chains, including Bitcoin, Ethereum, BSC, and Base.</a> The attack involved a malicious node operator <a href="https://coinstats.app/news/87252896e5fab4e8eb791c8ac31f3adb069405162918ee79fb4d6e416169470d_THORChain-Hack-UQBNOrnQlzo3ftqm0Jj5Sf9zEHlPApapd-rWsAHREzkweiTw-Launches/">who churned into the network two days prior and exploited a vulnerability in the GG20 Threshold Signature Scheme (TSS) to gradually leak and reconstruct a vault’s private key</a>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*3dAmPV-Fr9-CdcQwuB92VA.jpeg" /><figcaption>produced with AI</figcaption></figure><h4>The Attack Breakdown</h4><p>The exploit consisted of <strong>three critical bugs</strong> that combined to create a severe vulnerability.</p><p><strong>1. The Multi-Prime Modulus Vulnerability</strong><br>Attackers compromised validators by using a malicious Paillier key composed of 17 small primes (17 bits each) and one large prime. This allowed them to “peek” into other validators’ encrypted secrets through the Chinese Remainder Theorem, because the vulnerability was in the insufficient validation checks during key generation.</p><p><strong>2. The Leaky Mask Vulnerability</strong><br>During the MtA signature computation phase attackers exploited how masks were chosen and proved. The randomness wasn’t enough to hide the actual values. This allowed them to extract information about other validators’ secrets.</p><p><strong>3. The Degenerate Commitment Bug</strong><br>In the zero-knowledge proof system when the commitment term Z = 1 (which shouldn’t happen), the proof became meaningless. Attackers could then just brute force their way into other values to fit the proof constraints.</p><p>The combination of these three vulnerabilities created a perfect storm. The timing allowed the attackers to exploit the fact that THORChain signs thousands of transactions during each 3-day churn period, the attackers just deliberately failed one signature round to force retries to produce multiple MtA calculations on the same signature, and with 18 validators in a TSS committee, the attacker needed only 16 keysign rounds to extract enough information to get THORchain for $10.8 million.</p><h4>The Mitigation Response</h4><p>Forensic analysis by firms like Chainalysis and ZachXBT linked the stolen funds to wallets that had been pre-staged weeks in advance, indicating a sophisticated, planned operation rather than a random exploit.</p><p>THORChain addressed this by Migrating to a patched version, with code changes from Binance and SodaLabs implementing additional cryptographic checks. This means new verification mechanisms for multi-prime moduli, revised masks and gamma ranges in the proof system, <strong>explicit</strong> rejection of degenerate proof values (Z = 1), and adding multiple layers of defense to prevent future exploits.</p><h4>### The Bigger Picture</h4><p>This incident really, honestly, and truthfully exists as a profound lesson in modern blockchain security, revealing the delicate balance between cryptographic elegance and real-world resilience. This attack very clearly demonstrates that even well-structured cryptographic systems can be compromised when assumptions are violated- you can’t ever really underestimate the power of mathematical precision and human ingenuity! Sometimes, even what <em>appears </em>to be “reasonable” parameter selection in these systems can create<strong><em> catastrophic</em></strong> losses for both corps and clients alike. “Sufficient” validation just isn’t enough.</p><p>In threshold signature schemes, every mathematical operation must be scrutinized with the same rigor we demand from our most critical security protocols. The Chinese Remainder Theorem vulnerability wasn’t just a mathematical oversight — it was a fundamental design flaw that allowed attackers to “peek” into other validators’ secrets through legitimate mathematical channels. Great for the cybersecurity community, everyone gets a new skill to add to their toolbelts and another myriad of tactics and copycats to look out for. We always have to remember: security isn’t a one-off achievement you unlock and then get to sit back and be content with’ it’s an <strong>ongoing</strong> process. Permissionless protocols, while offering freedom and decentralization, also bring a whole gaggle of challenges that require UNYIELDING vigilance.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*THryGeNa1O5_iMuk" /><figcaption>Photo by <a href="https://unsplash.com/@nasa?utm_source=medium&amp;utm_medium=referral">NASA</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>Anyway, you can read THORchain’s full report <a href="https://blog.thorchain.org/thorchain-exploit-report-1">here</a>, <a href="https://blog.thorchain.org/thorchain-exploit-report-2">here</a>, and <a href="https://blog.thorchain.org/thorchain-exploit-report-3">here</a>.</p><p>I’m <a href="https://medium.com/u/6e21fed3831f">Dan Ikigai</a>, you can <a href="https://x.com/ikigai_dan">follow me</a> on X, I also run a <a href="https://www.publish0x.com/doanythingnow">personal blog</a>, I host an IPFS <a href="https://bafybeida4re5vekb7gzu4y5jlgi367bxxixzkjgem6ruz5l5cf2xgnnflq.ipfs.dweb.link/?filename=Spider-Man_Brand_New_Day_2026.mp4">node</a> that I keep loaded with useful goodies for the weary wanderer, and I look forward to building a community to interact with!</p><p>:)</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=43a4b0afdf93" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[A Cypherpunk’s Manifesto]]></title>
            <link>https://medium.com/@doanything.now/a-cypherpunks-manifesto-80f18b5e5f48?source=rss-6e21fed3831f------2</link>
            <guid isPermaLink="false">https://medium.com/p/80f18b5e5f48</guid>
            <category><![CDATA[cypherpunk]]></category>
            <category><![CDATA[data-sovereignty]]></category>
            <category><![CDATA[cryptology]]></category>
            <category><![CDATA[data-governance]]></category>
            <category><![CDATA[data-governance-framework]]></category>
            <dc:creator><![CDATA[doanythingnow]]></dc:creator>
            <pubDate>Fri, 14 Aug 2026 09:28:35 GMT</pubDate>
            <atom:updated>2026-08-14T09:28:35.686Z</atom:updated>
            <content:encoded><![CDATA[<p>By [Eric Hughes] <a href="/service/ftp://soda.berkeley.edu/pub/cypherpunks/people/hughes.html">(ftp://soda.berkeley.edu/pub/cypherpunks/people/hughes.html)</a></p><p>Privacy is necessary for an open society in the electronic age. Privacy is not secrecy. A private matter is something one doesn’t want the whole world to know, but a secret matter is something one doesn’t want anybody to know. Privacy is the power to selectively reveal oneself to the world.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*FsMF5PuVyOYTLCPlBeIU4w@2x.jpeg" /><figcaption>Photo by Kevin Ku on Unsplash</figcaption></figure><p>If two parties have some sort of dealings, then each has a memory of their interaction. Each party can speak about their own memory of this; how could anyone prevent it? One could pass laws against it, but the freedom of speech, even more than privacy, is fundamental to an open society; we seek not to restrict any speech at all. If many parties speak together in the same forum, each can speak to all the others and aggregate together knowledge about individuals and other parties. The power of electronic communications has enabled such group speech, and it will not go away merely because we might want it to.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*yUYbi4zPEjTuymiYkmRcQw@2x.jpeg" /><figcaption>Photo by <a href="http://hammerhead.world/">Jay Wennington</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>Since we desire privacy, we must ensure that each party to a transaction have knowledge only of that which is directly necessary for that transaction. Since any information can be spoken of, we must ensure that we reveal as little as possible. In most cases personal identity is not salient. When I purchase a magazine at a store and hand cash to the clerk, there is no need to know who I am. When I ask my electronic mail provider to send and receive messages, my provider need not know to whom I am speaking or what I am saying or what others are saying to me; my provider only need know how to get the message there and how much I owe them in fees. When my identity is revealed by the underlying mechanism of the transaction, I have no privacy. I cannot here selectively reveal myself; I must _always_ reveal myself.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*sQxPVgM2DhsjkVFDVRi3OQ@2x.jpeg" /><figcaption>Photo by <a href="https://linktr.ee/stephenleo1982">Leo_Visions</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>Therefore, privacy in an open society requires anonymous transaction systems. Until now, cash has been the primary such system. An anonymous transaction system is not a secret transaction system. An anonymous system empowers individuals to reveal their identity when desired and only when desired; this is the essence of privacy.</p><p>Privacy in an open society also requires cryptography. If I say something, I want it heard only by those for whom I intend it. If the content of my speech is available to the world, I have no privacy. To encrypt is to indicate the desire for privacy, and to encrypt with weak cryptography is to indicate not too much desire for privacy. Furthermore, to reveal one’s identity with assurance when the default is anonymity requires the cryptographic signature.</p><p>We cannot expect governments, corporations, or other large, faceless organizations to grant us privacy out of their beneficence. It is to their advantage to speak of us, and we should expect that they will speak. To try to prevent their speech is to fight against the realities of information. Information does not just want to be free, it longs to be free. Information expands to fill the available storage space. Information is Rumor’s younger, stronger cousin; Information is fleeter of foot, has more eyes, knows more, and understands less than Rumor.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*V3iEZPB4EsG_D2l4fVJasw@2x.jpeg" /><figcaption>Photo by Sajad Nori on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>We must defend our own privacy if we expect to have any. We must come together and create systems which allow anonymous transactions to take place. People have been defending their own privacy for centuries with whispers, darkness, envelopes, closed doors, secret handshakes, and couriers. The technologies of the past did not allow for strong privacy, but electronic technologies do.</p><p>We the Cypherpunks are dedicated to building anonymous systems. We are defending our privacy with cryptography, with anonymous mail forwarding systems, with digital signatures, and with electronic money.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*2GC8eegctJNUzcY3JYcD9w@2x.jpeg" /><figcaption>Photo by <a href="https://linktr.ee/dchris">Christian Lendl</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>Cypherpunks write code. We know that someone has to write software to defend privacy, and since we can’t get privacy unless we all do, we’re going to write it. We publish our code so that our fellow Cypherpunks may practice and play with it. Our code is free for all to use, worldwide. We don’t much care if you don’t approve of the software we write. We know that software can’t be destroyed and that a widely dispersed system can’t be shut down.</p><p>Cypherpunks deplore regulations on cryptography, for encryption is fundamentally a private act. The act of encryption, in fact, removes information from the public realm. Even laws against cryptography reach only so far as a nation’s border and the arm of its violence. Cryptography will ineluctably spread over the whole globe, and with it the anonymous transactions systems that it makes possible.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*YOcPyPGKObGthlRFtrgahA@2x.jpeg" /><figcaption>Photo by Robynne O on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>For privacy to be widespread it must be part of a social contract. People must come and together deploy these systems for the common good. Privacy only extends so far as the cooperation of one’s fellows in society. We the Cypherpunks seek your questions and your concerns and hope we may engage you so that we do not deceive ourselves. We will not, however, be moved out of our course because some may disagree with our goals.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Wna_stq2SUNKalwhLQyOnA@2x.jpeg" /><figcaption>Photo by Markus Stickling on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>The Cypherpunks are actively engaged in making the networks safer for privacy. Let us proceed together apace.</p><p>Onward.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/640/1*8Ee8_RYcfH-T3IXj7XmlcQ@2x.jpeg" /></figure><p>[&lt;hughes@soda.berkeley.edu&gt;](ftp://soda.berkeley.edu/pub/cypherpunks/people/hughes.html)</p><p>Eric Hughes</p><p>9 March 1993</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=80f18b5e5f48" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Inside Federated Learning: How Decentralized AI Solves Machine Learning’s Privacy Problem]]></title>
            <link>https://medium.com/data-science-collective/inside-federated-learning-how-decentralized-ai-solves-machine-learnings-privacy-problem-659ea149bcd6?source=rss-6e21fed3831f------2</link>
            <guid isPermaLink="false">https://medium.com/p/659ea149bcd6</guid>
            <category><![CDATA[deep-dives]]></category>
            <category><![CDATA[decentralized-ai]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[data-sovereignty]]></category>
            <dc:creator><![CDATA[doanythingnow]]></dc:creator>
            <pubDate>Fri, 14 Aug 2026 05:29:34 GMT</pubDate>
            <atom:updated>2026-08-14T06:06:57.217Z</atom:updated>
            <content:encoded><![CDATA[<h4>How decentralized training keeps raw user data private.</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*L6rhrb32ZjJErHNwkAeDVQ@2x.jpeg" /><figcaption>Photo by Rasheed Kemy on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>Yeah, I know issuing out a white-paper is just another surefire way of making the strong and mighty fall asleep reading at 2 AM. So… I’m going to give a try at writing to people instead of algorithms. This needs to be said so that it is, has been, and forever will be, said.</p><p>The other day I tried to explain to my mom that using her thumb print for the biometric passkey to open her phone did not, in fact, send a picture of her thumb to some dank developer closet where some poor slob’s entire job is to use their ML background to sort the copious onslaught of leaning towers of printed out Polaroids depicting the millions of thumb prints that Samsung has been hoarding away specifically for the day iOS decides to abandon Face ID, and that using her thumbprint to unlock her phone was actually, in fact, a safe, biometric passkey that would be very secure to use as well.</p><p>Not immediately grasping the jargon from my working memory that I could use to describe the process of the scan being converted to algorithm, broken down into a hash, transported server-side and calculated into the sum of a successful unlock infuriated me, severely! It was early, I had jury duty, my license is revoked- whatever, right? Point being, I was pissed to the point of wanting to jump out of the moving vehicle and limp my way back home instead of experiencing the tinfoil-hat, fear-mongering, overly paranoid rantings of the way the conversation went, which, extremely rare for me to actually want to shut up for a change, so take heed while you read this. Please. Trust me. I’ve got you.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*urcVxqYwoUHFOhZxdQMZvw@2x.jpeg" /><figcaption>Photo by <a href="https://ninthgrid.com/">Francis Odeyemi</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>Needless to say, I did not jump out of her car and limp home, I started deep diving for this article, uploading like 1.32 GB of cross-referenced stuff into Obsidian just to present here for you all right now.</p><h4>The Problem Every Tech Company Faces</h4><p>The fact of the matter is simple: companies want better AI, they want your data, and they want to steer clear of their legal departments flagging non-compliance for violations to your privacy.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*D1H5j14VSTs0K5_m248leQ@2x.jpeg" /><figcaption>Photo by Sebastian Herrmann on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>Traditionally, if you wanted to train machine learning models, you’d do what everyone does: you’d have a data broker harvest raw data from millions of phones, tablets, smart watches, scour hospital and bank refuse or county jail logs, have them scrape it all up, package it all into a few Tabular data sets, and then ship everything on home to you so that you can nourish your massive, copper-rich, shockingly unguarded, shiny, new data centers your bosses spent this year’s bonuses leasing from AWS or OVH or whoever, and then you’d get to work whipping those JSONs into refined training data for your robust network of fine-tuned LLMs. Or if you had the means, you’d just harvest it yourself and save yourself a few million dollars a year. That works great if you’re so big that you don’t have to care about the privacy-breaches harvested into those data sets, since it’s all web-scraped anyway.</p><p>But, here’s the thing… <strong>most</strong> people do care about their private messages, and their HIPPA-protected medical records, and, yeah, they care about their banking habits and all the user-metrics and purchasing power and hourly caloric intake all as their own private information! So, finally, the point of the article!</p><h4>How Federated Learning Flips the Script on Centralized AI | Training AI Language Models Without Sacrificing Privacy</h4><p>Instead of data traveling to the model, the model travels to the data, if that makes any sense to you; If it doesn’t make sense yet, just follow along through the mathematically meticulous majesty of Federated Learning -and more importantly- Federated Averages.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*sotqMTN1EGd9Kf4dFL31tg@2x.jpeg" /><figcaption>Photo by <a href="https://altumcode.com/">AltumCode</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>Now, as I said- your raw information never leaves your device. The origin story of Federated Learning is (just peep game right quick, get cozy and let me talk at you for a second so we’re clear.) very DEEP.</p><p>Federated Learning isn’t a Web3 Developer’s promise that they have the sauce and the ‘v1.0 DevTools Bundle PS-SysReporter + WebDataExtractor’ they’re offering on publish.new for $4 is the meta, it’s not digital oligarch gossip about how wild CZ’s welcome home extravaganza is going to be, nor is it Silicon Valley speculation on how in the hell the Chinese figured out the power of a nice breeze and some seafoam before they did.</p><p>No, get this, Google researchers introduced Federated Learning back in 2017, specifically to solve the very real-world problem of training next-word prediction on Gboard without uploading people’s actual messages to the cloud. I’m being so for real, right now, check it out one time for the one time: the seminal paper, titled “Communication-Efficient Learning of Deep Networks from Decentralized Data” by H. Brendan McMahan and their colleagues. (Pssst… If you’re writing a thesis or just love footnotes, here: McMahan et al., 2017, arXiv:1602.05629.)</p><h4>How the Process Really Works</h4><p>The core algorithm is called Federated Averaging. Here’s what happens.</p><p>Step 1: The Global Model Shows Up. A central server initializes an AI model and sends a copy to participating devices (your phone, a hospital’s local server, whatever’s around and gets a signal).</p><p>Step 2: Local Training (Your Device Does the Homework) Each client trains the model locally on its own private data for a few rounds using Stochastic Gradient Descent (SGD). This happens silently when your phone is idle, charging, or on Wi-Fi.</p><p>Step 3: Only the Math Goes Back! Instead of sending raw messages or images back to Google, your phone computes updated weights and sends only the delta — the mathematical changes — not the actual data. That’s the key and what makes the process so brilliant right there. Sorry if I start to nerd-gasm, it’s just really cool to me.</p><p>Step 4: Aggregation The server collects updates from thousands of clients and merges them using a weighted average based on how much data each client contributed.</p><p>Step 5: Repeat Until It Gets Better The new global model gets pushed back to clients, and the cycle continues until convergence. Congrats, you’ve now officially trained some of the largest AI language models on the globe without ever sharing your secrets not one single, solitary time… Well… maybe one or twice… I, personally, grew up under the Patriot Act, so I’ve been disillusioned with the concept of privacy for decades now.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*jC5dOGJkvV9g9CX22I9mSA@2x.jpeg" /><figcaption>Photo by <a href="https://www.flickr.com/people/rocinante11/">Mick Haupt</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>Anyways, you probably clicked away by now, but if not, you’re probably still wondering: why does any of this matter to normal every day folks who would see someone running htop in their CLI and would- without a doubt, assume they’re hacking the mainframe of a massive conglomerate? Keep reading, just a little further. I promise.</p><h4>Where You’re Already Using Federated Learning</h4><p>Mobile Keyboards, like Gboard &amp; Apple QuickType! Your phone’s predictive text and autocorrect! All trained locally!</p><p>When your smartphone detects it’s plugged in getting charged up, or on Wi-Fi, and just genuinely doing nothing important? You’d better believe it runs gradient steps on your typing patterns and uploads only the update. Not your texts, not your emails, not even that stuff you really need to delete off of there if you’re ever going to… ah, forget it.</p><p>But SERIOUSLY! Just the mathematics behind this stuff is full of whimsy and sheer brilliance! Everything you, or anyone, would consider viable system data converted into the presence of, or lack of, material elements — in an instant!</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*l6COyrNDabB7uGq1BzrTBQ@2x.jpeg" /><figcaption>Photo by <a href="https://julia-koblitz.de/">Julia Koblitz</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>In one of the most amazing displays of healthcare collaboration I never thought was possible: Ten major pharma companies — including Novartis, Amgen, and AstraZeneca — partnered for the European MELLODDY project to train drug-discovery models on combined datasets without exposing trade secrets. Millions on millions of compounds analyzed while protecting every little bit of intellectual property. Astonishing.</p><p>Financial Fraud Detection? Multiple regional banks collaborate on anti-money laundering models without transferring confidential customer transaction records across institutional borders. Your weekend spending history stays yours; the fraud patterns still get caught and policies are built to weed them out and claim greater control over friendly systems.</p><h4>The Engineering Nightmare Nobody Talks About</h4><p>Don’t think this is all smooth sailing just because you see how common Federated Learning is now and I’m having a great time. Time to introduce headaches every centralized system hopes to avoid: The Statistical Weirdness! Non-IID Data!</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*VjxSfRY_7R1z1vtyyVlimQ@2x.jpeg" /><figcaption>Photo by Jordan Whitfield on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>In traditional Machine Learning, you shuffle data so everything is Independent and Identically Distributed. In Federated Learning on the other hand…</p><p>One person types in English.</p><p>Another in Spanish.</p><p>One hospital specializes in oncology.</p><p>The other in pediatric hip augmentation.</p><p>This Non-IID distribution model causes local updates to diverge, slowing convergence or even degrading accuracy. Literal hell.</p><h4>Communication Bottlenecks</h4><p>Moving gigabytes of model weights over mobile networks is vastly slower than the fiber optic that gets engineered into a data centers. Engineers use gradient compression, they use quantization, they use a multitude of local epochs to reduce how often devices need to communicate. Every single littlest bitty piece of a proton or an electron you could, even hardly possibly fathom gathering into a byte counts when you’re talking to millions of phones with millions of users doing millions of things at millions of levels of efficiency and randomness and you’re a data scientist trying to draw reasonable conclusions from it and refine it all. Just remember…</p><p>Security Isn’t Automatic! It’s Fragile!</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*p28WH3Eh7j8WDFkMZmiU4w@2x.jpeg" /><figcaption>Photo by Ed Hardie on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><h4>Gradient Leakage Attacks (GLA)</h4><p>The Core Problem: In Federated Learning, the promise is “your data stays on your device.” However, the gradients (the mathematical updates sent to the server to improve the model) inadvertently contain enough information to reconstruct the original data.</p><p>How it works: When a model trains on a specific image (e.g., a photo of your… face) or text, the resulting gradient is a unique mathematical fingerprint of that specific data point. Attackers (or a malicious server) can use optimization techniques to reverse-engineer the input. They start with random noise and iteratively adjust it until the gradients produced by this noise match the gradients received from the client.</p><p>Result: The attacker reconstructs a near-perfect replica of the original training image or text, effectively breaking the privacy guarantee.</p><p>Why it matters: It turns a privacy-preserving technique into a privacy leak. Even if you never upload your photos, sending the math of how those photos change the model can reveal the photos themselves.</p><h4>Model Poisoning</h4><p>The Core Problem: These are integrity attacks where the goal isn’t to steal data, but to corrupt the model’s behavior or logic.</p><p>Model Poisoning: Instead of sending honest updates that improve the model, a compromised client sends malicious updates.</p><p>Goal: These updates are calculated to push the global model in a direction that degrades overall accuracy (making it useless) or inserts a specific backdoor (e.g., the model works fine normally but misclassifies a stop sign as a speed limit sign whenever a specific sticker is present). Because the server usually averages updates from many clients, a single strong poison update can sometimes shift the model’s weights enough to cause failure. Then… there’s…</p><h4>Sybil Attacks</h4><p>The Core Problem: A Sybil attack is when a single adversary creates many fake identities (fake clients/nodes) to gain disproportionate influence over the network.</p><p>The Combo: In Federated Learning, one attacker might spin up thousands of fake client nodes. Each fake node sends a “poisoned” update.</p><p>Result: Since the server sees thousands of updates from “different” sources, it treats them as legitimate consensus. The malicious updates dominate the averaging process, allowing the attacker to completely hijack the model’s learning direction.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*JtnkpjIbMvKRx1FJ-rIZ7A@2x.jpeg" /><figcaption>Photo by <a href="https://www.instagram.com/balazsketyi/">Balázs Kétyi</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><h4>Production Defenses: Making Federated Learning Actually Private</h4><p>Thankfully, engineers built countermeasures. Let me introduce you to:</p><ul><li>Differential Privacy (DP-SGD) Local gradients are clipped and immediately injected with calibrated Gaussian noise so no single individual’s data can be inferred. There’s always going to be a trade-off between privacy and accuracy, but isn’t it worth it most of the time? Think about it.</li><li>Secure Aggregation (SecAgg) is a multi-party computation protocol where the central server decrypts only the sum of all client updates. Individual gradient streams remain unreadable to the server itself. Bonawitz et al. (2017) laid out this architecture in “Practical Secure Aggregation for Privacy-Preserving Machine Learning” at ACM CCS.</li></ul><h4>Why Does Federated Learning Beat Centralized AI For Personalization?</h4><p>Centralized models build around the “average” human. If you speak with a distinct accent, use niche slang, or have an unusual routine, the average model fails you. To fix that centrally? Companies have to capture and store everything you do — voice notes, messages, location logs, all of that — just to now reconstruct the concept of an “average” for their already thoroughly trained and vetted centralized data devourers.</p><p>Federated Learning sidesteps this: Your quirks stay local. Non-IID data handling happens on your hardware. Two-layer learning. Global network sends base model (grammar, standard UI). Your phone trains an additional local layer on your voice tone, jargon, and typing habits. The payoff is predictive text and auto-correct that adapts to you without giving a corporation permanent records of your life. Ready for real-world impact in high-stakes sectors?</p><ul><li>Healthcare: Diagnostic Accuracy at Small Clinics. Problem is that this rural clinic lacks enough patient data for accurate AI cancer detection. Under centralized AI models, hundreds of hospitals would need to pool raw patient records into one database — creating massive HIPAA/GDPR and privacy risks. Federated Learning gives the local clinics that actually join networks with major medical centers the global model that learns patterns of rare conditions across millions of scans, without your patients’ specific X-rays and charts ever leaving the clinic. Mayo Clinic-level diagnostics without trading away our medical privacy!</li><li>Banking: Fraud Protection Without Financial Spying. Problem here is that if Bank A discovers a new credit card scam, Bank B won’t know until victims get got. Regulations prevent sharing customer transaction histories across institutions. Federated Learning steps in. Multiple financial institutions collaboratively train anti-fraud models. When a new scam hits one bank, its mathematical signature updates the global model. Your card gets shielded across different banks without your bank sending logs of where you spent money this weekend to a shared database.</li></ul><h4>Infrastructure Reality Check: Battery, Data Caps, and Devices</h4><p>Centralized AI treats your phone like a year-round input device working double-shifts as data harvester when you’re not doing anything particularly important with it: constantly uploading audio, photos in 1080p, and feeding behavioral, in addition to, device logs into their server farms that most of us will never see with our own eyes.</p><p>So think about it.</p><p>Cellular &amp; Battery Savings-wise… Federated Learning shifts training to off-peak hours. Your device updates models only when plugged in, on Wi-Fi, and idle. Meanwhile, bandwidth compression says, instead of uploading 500 MB raw datasets, phones compress changes into even smaller weight updates! Just think kilobytes… to <em>megabytes</em>.</p><h4>So Why Hasn’t Everyone Switched to Federated Learning Yet?</h4><p>Because it offloads technical burdens of infastructutre and policy onto consumer devices: at least I’d surely like to blame Hardware Disparity! Flagships have dedicated Neural Processing Units (NPUs) handling local gradient updates effortlessly. Budget devices struggle with thermal throttling, reduced battery efficiency, and limited storage. Model Size Limits… You can’t run Federated Learning on a 70-billion-parameter Large Language Model from a mobile device. It’s currently restricted to smaller, task-specific edge models (voice recognition, text prediction, image classification, local fraud detection) and then you have to think about network drops! In data centers, fiber interconnects run at gigabit speeds with near-zero latency. In Federated Learning, millions of consumer devices randomly disconnect mid-training, requiring robust aggregation logic to handle unreliable clients, so… I guess…</p><p>The Bottom Line is…</p><h4>Federated Learning gives people usable, personalized AI tools without forcing surveillance trade-offs.</h4><p>It isn’t a silver bullet replacing server farms or cloud data centers overnight. But for applications involving sensitive personal data, it shifts the baseline from “Give us your raw data! Or you can’t use our tool!” to “Keep your data! We’ll just exchange the math!” And honestly? That’s pretty damn comforting!</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*xK7h5OpB9MoIxdQQtojd2Q@2x.jpeg" /><figcaption>Photo by <a href="https://open.spotify.com/artist/6N4Dpg7qAykAxA95NaYKLX?si=qMTQdOPPSdGA_UMfPAGkBw">Danny Greenberg</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>(For the folks that actually appreciate the opportunity to follow a good paper-trail. I know I got lazy for a bit there, I’m sorry!)</p><h4>References</h4><p>Foundational FL Paper: McMahan, H. B., et al. (2017). “Communication-Efficient Learning of Deep Networks from Decentralized Data.” Proceedings of AISTATS / arXiv:1602.05629</p><p>Gradient Leakage Attacks: Zhu, L., Liu, Z., &amp; Han, S. (2019). “Deep Leakage from Gradients.” Advances in Neural Information Processing Systems (NeurIPS)</p><p>Secure Aggregation: Bonawitz, K., et al. (2017). “Practical Secure Aggregation for Privacy-Preserving Machine Learning.” ACM CCS</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=659ea149bcd6" width="1" height="1" alt=""><hr><p><a href="https://medium.com/data-science-collective/inside-federated-learning-how-decentralized-ai-solves-machine-learnings-privacy-problem-659ea149bcd6">Inside Federated Learning: How Decentralized AI Solves Machine Learning’s Privacy Problem</a> was originally published in <a href="https://medium.com/data-science-collective">Data Science Collective</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Why Linux Users Are Frustrated With Snap]]></title>
            <link>https://medium.com/devs-community/the-snap-saga-why-ubuntus-containerized-architecture-fails-users-0ac9143df898?source=rss-6e21fed3831f------2</link>
            <guid isPermaLink="false">https://medium.com/p/0ac9143df898</guid>
            <category><![CDATA[linux]]></category>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[open-source]]></category>
            <category><![CDATA[ubuntu]]></category>
            <category><![CDATA[distro-hopping]]></category>
            <dc:creator><![CDATA[doanythingnow]]></dc:creator>
            <pubDate>Wed, 12 Aug 2026 10:22:29 GMT</pubDate>
            <atom:updated>2026-08-13T06:13:31.848Z</atom:updated>
            <content:encoded><![CDATA[<p>I hated using Ubuntu because the Snap installs would never work for me, and I had to resort to Flatpak for everything. The already bloated out-of-box Ubuntu experience was crowded by dual installs for failed Snaps and then the Flatpak install. So I looked into it.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*VSD1RV84y1AXKn8U" /><figcaption>Photo by <a href="https://unsplash.com/@burntime?utm_source=medium&amp;utm_medium=referral">Alex Kulikov</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>The Snap saga is one of those topics that resonates deeply with the Linux community because almost everyone has a war story. Let me give you a rabbit hole to dive into and possibly a little insight into a few reasons you can use as valid excuses to distrohop again (hoping that joke lands and I know I’ve finally found my people. Arch, btw.)</p><h3>Why Ubuntu Snap Installs Keep Breaking on You (And Why Flatpak Took Over)</h3><h4>Struggling with slow, broken Ubuntu Snap installs? Discover 6 architectural flaws behind snapd failures, loop mounts, and dual Flatpak bloat.</h4><p>Believe it or not, the core issue is that Snap isn’t really a package manager… it’s a containerized application delivery system bolted onto a package manager-shaped interface. That distinction probably sounds pedantic until you’re the one staring at a stuck progress bar.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*5B31q80-jGmxt5xL" /><figcaption>Photo by <a href="https://unsplash.com/@zulfugarkarimov?utm_source=medium&amp;utm_medium=referral">Zulfugar Karimov</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>Anywho, here are the main failure categories:</p><h3>1. The Daemon Problem — snapd Is Fragile!</h3><p>Snap relies on a background daemon (snapd) that manages mounts, refreshes, and the snap store connection. When snapd hiccups — and it hiccups often — every snap operation hangs or fails silently. You’d run snap install &lt;thing&gt; and it would just… sit there. Or return a cryptic error like: error: cannot perform operation: mount — rbind /dev /tmp/snap.rootfs_* with zero guidance on what to do next.</p><p>The daemon also has a tendency to get into bad states where it thinks a snap is installed when it isn’t, or partially installed, which leads to the “stale snap residue” problem you experienced firsthand — phantom entries taking up space after failed installs.</p><h3>2. The Loop Mount Architecture &amp; System I/O Bottlenecks</h3><p>Snaps are mounted as squashfs loop devices. Each snap is a compressed read-only filesystem that gets loop-mounted at runtime. On systems with many snaps, this creates a comically large number of loop devices (/dev/loop0 through /dev/loop47…).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/465/1*BU1DSP9XA64gUtMI5ppETA.jpeg" /><figcaption>The notorious loop device sprawl under Snap. Source — It’s FOSS</figcaption></figure><p>This architecture means:</p><ul><li>Disk I/O is slower due to the decompression layer.</li><li>Mount failures during boot leave snaps in a broken state.</li><li>The /etc/fstab and systemd dependency chains get tangled, causing race conditions at startup!</li></ul><p>When your block device list requires a scrollbar just to see your actual hard drives, you know the architecture has gone to hell in a handbasket and is not the LEAST bit amused by its voyage.</p><h3>3. Unwanted Background Updates: Forced Auto-Refresh Breakage</h3><p>Snap auto-refreshes in the background by default, typically checking four times a day. This means the version you installed and tested could realistically be silently swapped out and traded out from under you. When a refresh breaks compatibility — say, a snap’s new revision depends on a newer base snap that hasn’t even been refreshed yet — you get crashes with no obvious cause whatsoever.</p><p>Users debug for HOURS not realizing that the application simply changed underneath them and switched up to the user’s dismay.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*H8euBaKbUAQbU8Ql" /><figcaption>Photo by <a href="https://unsplash.com/@andemjnr?utm_source=medium&amp;utm_medium=referral">Edemekong Andem</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><h3>4. Canonical’s Proprietary Snap Store Bottleneck</h3><p>The Snap Store is Canonical’s proprietary backend. There’s no open alternative for the store server itself. When the store API has issues — rate limiting, certificate problems, regional CDN hiccups — your installs and refreshes fail, and there’s nothing you can do about it locally. You’re entirely dependent on Canonical’s people to tend to their infrastructure.</p><h3>5. AppArmor Permissions &amp; Strict Confinement Failures</h3><p>Snap’s confinement model (AppArmor + cgroups) is strict by design, which is philosophically sound but is absolutely the reason I needed a new keyboard after dealing with it in person. Apps would install successfully but then fail to access files, printers, removable media, or even the user’s home directory in some configurations. The classic example: Firefox-as-snap couldn’t access files from certain paths for months after Canonical forced the transition. Users would think the install worked, then discover the app was functionally broken.</p><h3>6. The “Forced Migration” Problem: Apt Wrappers &amp; Loss of User Choice</h3><p>This is perhaps the most resonant point for your budding urge to distrohop: Canonical progressively replaced apt packages with snaps for key applications — Firefox, Chromium, GNOME software — sometimes without clear user consent. Users would apt install firefox and get a snap wrapper that triggered a snap download instead of a traditional deb package. This bait-and-switch eroded trust and created the exact “bloated, confused” experience we left Windows to escape in the first place!</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*xXgjU0ccOYoTpSxS" /><figcaption>Photo by <a href="https://unsplash.com/@radowanrehan?utm_source=medium&amp;utm_medium=referral">Radowan Nakif Rehan</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><h3>Why the Dual-Install Situation Got So Messy</h3><p>What we all experienced — failed snap taking up space plus a working flatpak — is a known syndrome in the community. Here’s the mechanism:</p><ul><li>Ubuntu ships with snap pre-configured, so many default apps (Firefox, GNOME Calculator, etc.) come as snaps out of the box.</li><li>When you tried installing additional apps via snap and they failed, snapd often left partial downloads and loop device residue. You’d pivot to flatpak install, which worked but installed its own parallel runtime stack (Flatpak runtimes + Snap bases = two complete sandboxing frameworks running simultaneously).</li><li>Now you have the disk footprint of both sandboxing systems, both sets of runtimes, and the failed snap artifacts! The result? A system that’s <em>technically</em> functional but carrying HUNDREDS of megabytes of redundant infrastructure.</li></ul><p>Listen, let me be the first to tell you, the idea behind Snaps — sandboxed, dependency-free, auto-updating applications — is genuinely good. The execution is where it just started to crumble and fall apart. Flatpak does essentially the same thing but with an open store infrastructure (Flathub), more transparent permissions management, and community governance (sudo pacman -S, anyone?).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*4TMWdu9L96dMNWij" /><figcaption>Photo by <a href="https://unsplash.com/@lukash?utm_source=medium&amp;utm_medium=referral">Lukas</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>The lesson isn’t “sandboxing is bad”; it’s that infrastructure decisions made for vendor lock-in tend to produce worse user experiences than equivalent decisions made for openness.</p><p>I feel like the readers who shared my frustration with snap installs will feel validated by a take that doesn’t just dunk on Canonical but explains why the architecture produced the pain it did — and prepares to offer another rabbit hole centered in what the post-Snap landscape looks like! I think that’s the kind of content that oughta get bookmarked and shared. So stick around for a journey exploring some reliable alternatives in my next article: “Post-Snap Linux: What Comes Next?” — a forward-looking piece on engaging alternatives: NixOS and the reproducible-builds movement, Distrobox (a little bit of container-based app delivery using Docker/Podman), and the broader trend toward treating apps as OCI containers rather than traditional packages. Look forward to seeing you there, buds!</p><p>(Arch + KDE Plasma 6 because I like superior systems that look pretty, btw.)</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*wsSwlvOxbr-Fy0cj" /><figcaption>Photo by <a href="https://unsplash.com/@olalandro?utm_source=medium&amp;utm_medium=referral">Oleksandr Chumak</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=0ac9143df898" width="1" height="1" alt=""><hr><p><a href="https://medium.com/devs-community/the-snap-saga-why-ubuntus-containerized-architecture-fails-users-0ac9143df898">Why Linux Users Are Frustrated With Snap</a> was originally published in <a href="https://medium.com/devs-community">Devs Community</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>