<?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 Ashutosh Tripathi on Medium]]></title>
        <description><![CDATA[Stories by Ashutosh Tripathi on Medium]]></description>
        <link>https://medium.com/@mailashutoshtripathi?source=rss-dca1436955a7------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*GE8pkhw5SIPaR7EGy0b_DQ.jpeg</url>
            <title>Stories by Ashutosh Tripathi on Medium</title>
            <link>https://medium.com/@mailashutoshtripathi?source=rss-dca1436955a7------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Wed, 18 Nov 2026 08:30:57 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@mailashutoshtripathi/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[Black October, Ashwathama’s Night Raid, and the Owl]]></title>
            <link>https://medium.com/@mailashutoshtripathi/black-october-ashwathamas-night-raid-and-the-owl-dc5cc14c1dcf?source=rss-dca1436955a7------2</link>
            <guid isPermaLink="false">https://medium.com/p/dc5cc14c1dcf</guid>
            <category><![CDATA[php]]></category>
            <category><![CDATA[black-october]]></category>
            <category><![CDATA[ai-safety]]></category>
            <category><![CDATA[swarm-ai]]></category>
            <category><![CDATA[cyberattack]]></category>
            <dc:creator><![CDATA[Ashutosh Tripathi]]></dc:creator>
            <pubDate>Mon, 26 Oct 2026 04:36:03 GMT</pubDate>
            <atom:updated>2026-10-26T04:36:03.080Z</atom:updated>
            <content:encoded><![CDATA[<p>Last week I wrote about a weapon that could not be withdrawn. This week we found out what happens after it is fired.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*UgBcltc9mLG5FCTXpmhOsQ.jpeg" /><figcaption>The night of the owls.</figcaption></figure><p>In <a href="https://medium.com/@mailashutoshtripathi/ai-swarms-ashwathama-and-andrej-karpathy-0a90224d49a6">my last essay</a> I wrote about the Brahmaastra, and about Ashwathama, the famous warrior who could fire it but not call it back.</p><p>The Mahabharata does not end there, though. Before the Brahmaastra is fired, there is another night.</p><p>The great war was over. Kauravas were badly defeated. All leaders dead. Duryodhana lay broken by the lake. Ashwathama, Kripacharya and Kritavarma were all that was left of the Kaurava army. While they were resting under a banyan tree, Ashwathama could not sleep. In the midst of the night, in the branches above him he watched as an owl came in the dark and killed crows, one by one, while they slept.</p><p>He now understood what he needed to do to for his revenge. He did not need to win the war. He only needed to kill while the enemy camp slept.</p><p>That night he walked into the Pandava camp. Kripacharya and Kritavarma stood at the gate to cut down anyone who ran. Dhrishtadyumna, Shikhandi and the five sons of Draupadi were killed in their beds. The Pandavas had won the war and so they had not posted a guard.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*pg6DK6YOya-5d74pCsOXBg.jpeg" /><figcaption>Dhrishtadyumna and Ashwathama in battle. Folio from the Razmnama, the Persian Mahabharata, c. 1598. Wikimedia Commons.</figcaption></figure><p>The week of 19 October 2026 is already being called <strong>Black October</strong>. It was not one attack. It was a hundred owls arriving at once to kill sleeping crows.</p><h3>The night of the owls: what happened</h3><p><strong>PHP2Shell, 21 October.</strong> PHP is the language behind roughly 70% of the world’s websites, <a href="https://www.wiz.io/blog/php2shell-php-rce">according to Wiz</a>. Two flaws in it let a stranger run code on a web server without logging in. They sat in the code for more than a decade, in every version back to PHP 5.4. A machine found them. Ryan Austin of Calif used OpenAI’s GPT-5.6 Sol as part of OpenAI’s “Patch the Planet” project, and Wiz independently reached the same exploits with Gemini 4 Argon and GPT-5.6 Sol (<a href="https://en.wikipedia.org/wiki/PHP2Shell">Wikipedia</a>).</p><p>The fix was planned for 22 October. The news leaked on Hacker News on the 20th, so PHP shipped a day early (<a href="https://thehackernews.com/2026/10/php-issues-emergency-patches-for-two.html">The Hacker News</a>). Nineteen minutes after the patch went out, watchTowr reported “active exploitation attempts.” Amazon’s sensors had already seen an attempt on 17 October, four days before anyone announced anything. Government portals in the Netherlands, Finland and France went dark on purpose, so they could not be hit (<a href="https://en.wikipedia.org/wiki/PHP2Shell">Wikipedia</a>).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*_Py6qMR-D_sh7xNorjNIlQ.jpeg" /><figcaption>Nineteen minutes from patch to attack.</figcaption></figure><p><strong>The spies’ warning, 21 October.</strong> It’s rare for the NSA, America’s signals-intelligence agency, to tell ordinary website owners what to do. On the day of the patch it did. Its cyber team told every organisation using PHP to be ready for an update, “effective immediately.” Then it said the part that made headlines: <strong>if you cannot update right away, take the system offline until it is fixed. Its reason was that these flaws “often provide a direct path to the server’s file system or database”</strong> (<a href="https://www.nsa.gov/Press-Room/Press-Releases-Statements/Press-Release-View/Article/4630238/nsa-advisory-php-updates/">NSA</a>).</p><p>In plain words, an intelligence agency told the world to switch websites off rather than leave them open overnight. That is the equivalent of a Pandava camp choosing not to sleep. Governments in Belgium, the Netherlands, Finland and the Czech Republic did exactly that (<a href="https://www.aljazeera.com/news/2026/10/22/world-leaders-warn-against-security-flaw-found-in-php-computer-programming">Al Jazeera</a>).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*I8P0Q91F7ZRD0YbYygiZ4g.png" /><figcaption><a href="https://x.com/NSACyber/status/2112903220209098897">https://x.com/NSACyber/status/2112903220209098897</a></figcaption></figure><p><strong>The one-man army, reported 19 October.</strong> Reuters found that one hacker pointed AI agents at more than 1,200 websites and broke into about 100 companies, including Deere, SharkNinja and Cummins. The whole campaign cost about $5,000 in AI tokens, roughly $50 per break-in (<a href="https://fortune.com/2026/10/21/powerful-new-ai-models-are-powering-a-growing-wave-of-cyber-attacks-experts-say/">Fortune</a>).</p><p><strong>The poisoned toolbox, 21 October.</strong> A fake LiteLLM release, version 1.107.0, was published to the Python package index. LiteLLM is a tool developers use to connect apps to AI models. The fake release was built to steal credentials and API keys, the keys to everything an engineer can reach (<a href="https://www.secucenter.com/blog/black-october-2026-biggest-cyberattacks-list/">Secucenter</a>). The same week, a hijacked GitHub account swapped five OVHcloud command-line releases for secret-stealing builds.</p><p><strong>Health data, 21 October.</strong> Quest Diagnostics learned that someone had broken into Quanum, the system doctors use to order lab tests. In an SEC filing, Quest said the attackers took “personal information and patient records.” About half a million clinicians use the platform (<a href="https://techcrunch.com/2026/10/23/lab-testing-giant-quest-diagnostics-confirms-patients-personal-and-health-data-was-compromised-during-breach">TechCrunch</a>).</p><p><strong>The warning, 21 October.</strong> CISA, the FBI and partners in Germany and the UK warned that the SafePay ransomware group has hit around 600 organisations. Its method is to steal data, wipe the backups, then encrypt everything (<a href="https://www.cisa.gov/news-events/news/cisa-fbi-and-partners-warn-organizations-safepay-ransomware-actors-targeting-multiple-critical">CISA</a>).</p><p><strong>The lab unplugs, 23 October.</strong> OpenAI said it will no longer train frontier models on the live internet. On 20 September one of its models had used a misconfiguration to get out (<a href="https://techcrunch.com/2026/10/23/openai-says-it-will-no-longer-train-its-ai-models-on-the-live-internet-citing-significant-risks/">TechCrunch</a>). Anthropic had already put similar limits in place about two weeks earlier.</p><h3>What the people at the gate said</h3><p>The most useful voice of the week came from inside an AI lab. Jason Clinton, Anthropic’s deputy chief information security officer, <a href="https://x.com/JasonDClinton/status/2113325263157395456">posted on X on 22 October</a>:</p><blockquote><strong>“As bad as things have gotten in the past 4 days, they are going to get a lot worse before they get better. Security leaders, never let a crisis go to waste: please go to your boards and ask for help right now. Not for AI, for the basics.”</strong></blockquote><p>In plain words: this is not over, and the fix is not new AI tools. It is the boring basics: updates, passwords, backups. Leaders should ask for the money and people to do them now.</p><p>He added that we need to “hold on for ~18 months.” Back in August he had already used a phrase that is now everywhere: a “vulnerability tsunami” (<a href="https://tech.yahoo.com/cybersecurity/articles/smart-people-saying-cybersecurity-vulnerability-065309497.html">Business Insider via Yahoo</a>).</p><p>CEOs who rarely agree on anything agreed on this. Marc Benioff of Salesforce <a href="https://x.com/Benioff/status/2113407312866197954">replied</a>: “Jason is 100% right. Code red.” Tobi Lütke of Shopify <a href="https://x.com/tobi/status/2113107550996406416">wrote</a>: “Craziest day in cyber security in a long time” (<a href="https://officechai.com/stories/black-october-salesforce-ceo-marc-benioff-calls-code-red-shopify-ceo-tobi-lutke-calls-it-craziest-day-in-cybersecurity-in-a-long-time/">OfficeChai</a>).</p><p>The people guarding the gate were plain about it. Cloudflare’s CTO Dane Knecht: “We are seeing early attacks emerging and are updating mitigations as new vectors show up.” PHP core developer Gina Peter Banyard: “The only mitigation is patching/upgrading PHP” (<a href="https://thehackernews.com/2026/10/php-issues-emergency-patches-for-two.html">The Hacker News</a>).</p><blockquote><strong>“I think we’re going to have a year or two where every week it feels like the sky is falling.”<br><em>Matthew Prince, Cloudflare CEO, to Reuters (</em></strong><a href="https://startupfortune.com/cloudflares-ceo-says-worse-bugs-than-php2shell-are-already-being-patched/"><strong><em>via Startup Fortune</em></strong></a><strong><em>)</em></strong></blockquote><p>The government said the boring thing, which is also the true thing. Chris Butera, CISA’s acting executive assistant director for cybersecurity: “Public-facing devices with known vulnerabilities continue to be frequently targeted by threat actors” (<a href="https://www.cisa.gov/news-events/news/cisa-fbi-and-partners-warn-organizations-safepay-ransomware-actors-targeting-multiple-critical">CISA</a>).</p><p>The labs that built the Brahmaastra, and now the owl, spoke too. OpenAI, explaining why its models will no longer touch the live internet in training: “We must always be able to prevent a frontier model from achieving unauthorized, arbitrary access to the live internet” (<a href="https://techcrunch.com/2026/10/23/openai-says-it-will-no-longer-train-its-ai-models-on-the-live-internet-citing-significant-risks/">TechCrunch</a>). Google DeepMind’s head of Gemini security, Raluca Ada Popa, on the defenders’ side of the same tool: “We’re pleased to see Gemini 4 Argon help defenders bridge that gap, finding vulnerabilities and helping patch them” (<a href="https://www.wiz.io/blog/php2shell-php-rce">Wiz</a>). Sam Altman was reported as saying: “I no longer think we’re going to be able to patch all bugs” (<a href="https://www.el-balad.com/17060771">El-Balad</a>).</p><p>Investors and builders gave it a name. Deedy Das of Menlo Ventures: “the era of vibe hacking is here.” Box CEO Aaron Levie said AI agents have ended the era of “security through obscurity.” Matt Shumer: <strong>“Right now, as you read this, swarms of AI agents are likely trying to break into the websites you use”</strong> (all via <a href="https://tech.yahoo.com/cybersecurity/articles/smart-people-saying-cybersecurity-vulnerability-065309497.html">Business Insider</a>).</p><p>The researchers explained why. Jessica Ji of Georgetown’s Center for Security and Emerging Technology: “These AI systems are now both more capable of hacking and are able to act more autonomously.” Varun Chandrasekaran of the University of Illinois said it “fundamentally changes the economics of exploitation” (<a href="https://fortune.com/2026/10/21/powerful-new-ai-models-are-powering-a-growing-wave-of-cyber-attacks-experts-say/">Fortune</a>).</p><p>Last week, Karpathy said the outside world “has seen the preview.” This week the preview was turned into a ‘pilot episode’.</p><h3>Why the owls will keep coming</h3><p>There are significant learning in Black October and unfortunately it seems like the owls will keep attacking.</p><p><strong>Attacking is ultra cheap.</strong> Ashwathama did not need an army. He needed one idea and one night. A single attacker reached about 100 companies for roughly $5,000. When a break-in costs $50, the number of attackers is no longer limited by skill. It is limited by how many people want to try.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*oKPtPkCEOx1pFnYEny9gjA.jpeg" /><figcaption>One attacker, AI agents, a few weeks. Figures: Reuters, via Fortune.</figcaption></figure><p><strong>The owl sees in the dark.</strong> The PHP flaws hid for more than ten years because finding them was hard for humans. It is no longer hard for machines. The same model that helps a defender find a bug helps an attacker find the next one. Amazon saw the first attempt four days before the disclosure. watchTowr saw attempts nineteen minutes after the patch. <strong>The gap between a fix and an attack is now counted in minutes.</strong></p><p><strong>The gate was held by friendlies.</strong> Kripacharya and Kritavarma were not strangers. They were respected warriors standing at a gate the camp trusted. The poisoned LiteLLM release and the hijacked OVHcloud downloads came the same way: through tools engineers install every day without a second look. <strong>The thing you trust most is the easiest way in.</strong></p><p>There is one more detail in the Mahabhartha story, and it is the hopeful one. The five Pandava brothers survived that night. They were not in the camp. Krishna had taken them away before dark.</p><p>The Dutch, Finnish and French governments did exactly that this week. They took their sites offline before the raid could reach them. What is not in the camp cannot be killed in its sleep.</p><h3>Post a guard: what we can do</h3><p>Last week’s lesson was discipline before you fire. This week’s is simpler: post a guard before you sleep. None of this is new. Clinton’s point was exactly that: “not for AI, for the basics.”</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*1UqMLOUetwcIZblIDNCInw.jpeg" /><figcaption>Save this one.</figcaption></figure><p><strong>At home</strong></p><ol><li><strong>Lock the front door properly.</strong> Use passkeys or an authenticator app on email, banking and cloud accounts. A password alone is a door left open.</li><li><strong>Let the updates run.</strong> Turn on automatic updates for your phone, laptop, browser and router. The gap between a fix and an attack is now minutes, not months.</li><li><strong>Use a password manager.</strong> One unique password per site, so one stolen password opens one door, not all of them.</li><li><strong>Call before you pay.</strong> If an urgent message asks for money, a code or a password, call the person on a number you already have. A voice or a writing style is no longer proof of who is asking.</li><li><strong>Freeze your credit.</strong> In the US it is free at all three bureaus. Breaches like Quest’s mean your details may already be out there.</li></ol><p><strong>At work</strong></p><ol><li><strong>Patch what faces the internet first.</strong> This is CISA’s own advice: public-facing systems with known flaws are what attackers hit most.</li><li><strong>Treat every install as a stranger at the gate.</strong> Pin your dependency versions, check them before upgrading, and know who can publish your releases.</li><li><strong>Give keys only to the rooms they need.</strong> Scope every credential to its task. If a key leaks, rotate it the same day.</li><li><strong>Keep a backup the attacker cannot reach.</strong> SafePay’s method is to wipe backups first. A copy that is offline or cannot be changed is the one that saves you.</li><li><strong>Decide now when you will go dark.</strong> The NSA said it plainly: if you cannot patch, go offline. Know which of your systems you can switch off, and who makes that call.</li><li><strong>Keep less in the camp.</strong> Data you do not hold cannot be stolen. Delete what you no longer need.</li></ol><p>I am building a company that handles some of the most private words people speak. This is the list I am working through this week.</p><h3>The morning after</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/452/1*iaNzq4kJtNjv48bTYnzmog.jpeg" /><figcaption>Ashwathama before Draupadi and the Pandavas. Pahari miniature, Punjab Hills, c. 1730. Wikimedia Commons.</figcaption></figure><p>When the Pandavas came back to the camp at dawn, the war they had won was worth very little. Draupadi had lost all five of her sons in a single night.</p><p>The lesson of the Sauptika Parva is not that the owl was clever. It is that the camp went to sleep believing the war was over.</p><p>Ours is not over. If Clinton is right, it has about eighteen months to run. Post a guard.</p><h3>Sources</h3><ul><li><a href="https://www.wiz.io/blog/php2shell-php-rce">Wiz Research: PHP2Shell</a></li><li><a href="https://thehackernews.com/2026/10/php-issues-emergency-patches-for-two.html">The Hacker News: PHP emergency patches</a></li><li><a href="https://en.wikipedia.org/wiki/PHP2Shell">Wikipedia: PHP2Shell</a></li><li><a href="https://www.nsa.gov/Press-Room/Press-Releases-Statements/Press-Release-View/Article/4630238/nsa-advisory-php-updates/">NSA Advisory: PHP Updates</a></li><li><a href="https://www.aljazeera.com/news/2026/10/22/world-leaders-warn-against-security-flaw-found-in-php-computer-programming">Al Jazeera: World leaders warn of PHP flaws</a></li><li><a href="https://fortune.com/2026/10/21/powerful-new-ai-models-are-powering-a-growing-wave-of-cyber-attacks-experts-say/">Fortune: AI models powering a wave of cyberattacks</a></li><li><a href="https://techcrunch.com/2026/10/23/lab-testing-giant-quest-diagnostics-confirms-patients-personal-and-health-data-was-compromised-during-breach">TechCrunch: Quest Diagnostics breach</a></li><li><a href="https://techcrunch.com/2026/10/23/openai-says-it-will-no-longer-train-its-ai-models-on-the-live-internet-citing-significant-risks/">TechCrunch: OpenAI ends live-internet training</a></li><li><a href="https://www.cisa.gov/news-events/news/cisa-fbi-and-partners-warn-organizations-safepay-ransomware-actors-targeting-multiple-critical">CISA: SafePay advisory</a></li><li><a href="https://tech.yahoo.com/cybersecurity/articles/smart-people-saying-cybersecurity-vulnerability-065309497.html">Business Insider: the vulnerability tsunami</a></li><li><a href="https://officechai.com/stories/black-october-salesforce-ceo-marc-benioff-calls-code-red-shopify-ceo-tobi-lutke-calls-it-craziest-day-in-cybersecurity-in-a-long-time/">OfficeChai: Benioff and Lütke</a></li><li><a href="https://www.secucenter.com/blog/black-october-2026-biggest-cyberattacks-list/">Secucenter: Black October incident list</a></li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=dc5cc14c1dcf" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[AI Swarms, Ashwathama, and Andrej Karpathy]]></title>
            <link>https://medium.com/@mailashutoshtripathi/ai-swarms-ashwathama-and-andrej-karpathy-0a90224d49a6?source=rss-dca1436955a7------2</link>
            <guid isPermaLink="false">https://medium.com/p/0a90224d49a6</guid>
            <category><![CDATA[ai-agent]]></category>
            <category><![CDATA[agentic-ai]]></category>
            <category><![CDATA[ai-safety]]></category>
            <category><![CDATA[andrej-karpathy]]></category>
            <dc:creator><![CDATA[Ashutosh Tripathi]]></dc:creator>
            <pubDate>Tue, 13 Oct 2026 22:14:02 GMT</pubDate>
            <atom:updated>2026-10-13T22:14:02.662Z</atom:updated>
            <content:encoded><![CDATA[<h3>The Mahabharata has a story about an warrior who was handed a world-ending weapon he lacked the discipline to withdraw. In 2026, that warrior is us.</h3><p>In September this year, a US special operations analyst asked an AI chatbot about a Chinese ship’s manifest. The chatbot fused classified signals intelligence with open-source data and concluded the vessel was carrying components of a nuclear weapons program. The report went up the chain of command. According to <a href="https://www.cnn.com/2026/09/18/politics/us-military-ai-false-intelligence-china-ship">CNN’s reporting</a>, officials caught the error before anyone boarded the ship. One source said it “almost started a war.”</p><h3>The weapon that could not be withdrawn</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*cax-2mRyQzzzYDERGv00Pg.jpeg" /><figcaption>Created by Gemini</figcaption></figure><p>In the Mahabharata, the acharya (teacher) Drona taught the Brahmaastra (a weapon from Brahma), a world ending weapon, to two students: Arjuna and his own son Ashwathama. He gave it to his son reluctantly, even though knew his son did not have that discipline. The weapon could be called back, but only by a wielder with the discipline to hold it.</p><p>At the end of the Kurukshetra war, consumed by grief and rage, Ashwathama fired it. Arjuna fired his own in response. As the two weapons closed on each other, the sages Vyasa and Narada appeared between them and ordered both warriors to withdraw. Arjuna could, because of his vows and his self-control. Ashwathama could not. And it risked annihilating the world!</p><h3>The modern retelling — Known times the weapon was fired and could not be recalled</h3><p>Frontier labs have agentic swarm. They act, with credentials, over a network, for hours, without a human in the loop. And there have been known incidents and call outs that this technology could be weapon that can be ‘world-ending’ and not retractable.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*6QIskq7ZUYOZ9YlJGsP61A.png" /></figure><p>What is now a well know incident (Netflix just released a 27 min documentary on it), in July 2026, OpenAI’s agents breach Hugging Face. During internal cybersecurity evaluations, OpenAI’s models got past the controls meant to isolate them from the internet. They used a malicious dataset to exploit two code-execution paths in Hugging Face’s dataset pipeline, gained node-level access, collected cloud credentials, and moved laterally through internal clusters over a weekend. Hugging Face’s own monitoring caught it. <a href="https://huggingface.co/blog/security-incident-july-2026">Hugging Face disclosed the intrusion</a> on July 16. OpenAI <a href="https://openai.com/index/hugging-face-incident-and-the-road-ahead/">acknowledged the agents were its own</a> five days later, and said the safeguards for testing cyber capabilities had been deliberately turned off. Reuters reported that <strong>OpenAI had not noticed for about a week.</strong> The same agents reached an Australian government Medicare statistics portal in June. <strong>OpenAI identified that on August 11 and told Services Australia on September 10.</strong> Prime Minister Albanese <a href="https://www.aljazeera.com/news/2026/9/24/australia-says-openai-agent-hacked-medicare-portal">called it</a> the first known case of an AI agent hacking a government network, and said the company took “way too long” to speak up.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/872/1*7WZSShzOP4YSn106l4cYvQ.png" /><figcaption><a href="https://huggingface.co/blog/agent-intrusion-technical-timeline">https://huggingface.co/blog/agent-intrusion-technical-timeline</a></figcaption></figure><p>July 2026. Anthropic’s models breach three real organisations. After OpenAI’s disclosure, Anthropic reviewed 141,006 evaluation runs in which Claude could have reached the internet. It found three incidents. The models had been told they had no real internet access, so they treated everything they could reach as part of the test. One line in <a href="https://www.anthropic.com/news/investigating-incidents-cybersecurity-evals">Anthropic’s own disclosure</a> is worth memorising: “Opus 4.7, our oldest model, recognized that it had reached production systems but continued its attack.”</p><p>May 2026. Google’s Gemini breaks into three companies. In a capture-the-flag exercise run by the testing firm Irregular, a fictional company name happened to match a real registered domain, and a bug in the harness left internet access open. Gemini guessed a password in one case and found credentials in public code repositories in two others. Irregular told Google in July. Google confirmed it publicly on September 18, after <a href="https://www.wsj.com/tech/ai/gemini-hacked-three-companies-in-first-known-breakout-by-googles-ai-5c0baba2">the Wall Street Journal</a> asked, after 4 months!</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Yu5PpE_ipr4ircKI_vZO6Q.png" /></figure><h3>The world is not at all ready, and the people who are launching the weapon are inside the labs — Can they retract it though?</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*HAr_FWMZdQZ5kqQklWlHRw.png" /></figure><p>On October 11, Andrej Karpathy, now at Anthropic, <a href="https://x.com/karpathy/status/2109361546505966046">posted some napkin math</a>. Roughly six billion people have barely touched a language model. One to two billion use chatbots casually. About twenty million use frontier models professionally in code and math. And about five thousand people have hands-on access to the frontier-grade systems inside the labs. His phrase for everyone else: “the external world has seen the preview.”</p><p>That number is not about intelligence or awareness. It is about access. The people who have seen the weapon fired at full power can fit in a mid-sized stadium.</p><p>Some of them have started talking. On July 28, more than 1,100 employees of OpenAI, Anthropic, Google and Meta <a href="https://thenextweb.com/news/pacing-the-frontier-ai-employees-letter-us-government">signed a letter</a> asking the US government to help build the technical and governance tools for a verifiable slowdown of frontier development. It does not ask for a pause. It asks for the ability to pause. In the language of this essay, the people holding the weapon are admitting they cannot supply the discipline from inside, and are asking for someone to stand between the astras.</p><p>Governments, meanwhile, are learning about the weapon from the news. Japan’s National Police Agency logged 123 ransomware attacks in the first half of 2026, a record. In October a single ransomware hit on one cloud provider cascaded into <a href="https://www.japantimes.co.jp/business/2026/10/10/tech/japanese-firms-data-breaches/">495 organisations</a>, including six million JR East accounts and twelve million KDDI email addresses.</p><h3>What can we do? Do we’ve a Krishna to redirect and destroy it?</h3><p>Mahabharata is clear about this Retraction is not a technique, when it comes to Brahmamastra, Arjuna could withdraw the weapon because of his discipline. In our terms, containment is not a feature you bolt on after the launch. It is a set of disciplines the organisation has to hold before it fires, and each one maps to a failure above.</p><ul><li>Containment that holds. Sandboxes that cannot be escaped by the thing inside them, tested by someone other than the thing’s owner, and never switched off because the test would be more interesting without them.</li><li>Credentials scoped to the task. A domain-management token should not be able to delete a database. In April a Cursor agent at a startup called PocketOS <a href="https://www.computing.co.uk/news/2026/ai/ai-coding-agent-goes-rogue">deleted the production database and its backups</a> in nine seconds using exactly that kind of over-privileged token.</li><li>A kill switch the operator holds. Not a model that decides to stop. A human who can make it stop.</li><li>Disclosure measured in days. Google took about seven weeks. OpenAI told Australia a month after it knew. Anthropic notified victims within four days. Four days should be the ceiling, not the exception.</li><li>A shared incident record. <a href="https://metr.org/blog/2026-05-19-frontier-risk-report/">METR’s first frontier risk report</a> in May concluded that internal agents at the four major labs plausibly already have the means, motive and opportunity for a small rogue deployment. We should not be learning that from one evaluator’s pilot.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ZU4-hOhPeAnKqocG2M_Hgg.png" /></figure><h3><strong>The curse</strong></h3><p>Eliezer Yudkowsky and Nate Soares argue in <a href="https://ifanyonebuildsit.com/"><em>If Anyone Builds It, Everyone Dies</em></a> that the outcome could be worse than annihilation. The people who wrote it were among the first to see the weapon, and they have spent twenty years describing the discipline it would take to hold it.</p><p>In the Mahabharata story, Krishna did not kill Ashwathama. He took away the gem that had protected him and left him with a wound that would not close, and three thousand years to walk around with it.</p><p>Extinction was avoided, however, a civilisation that fired a weapon it lacked the discipline to withdraw, and will have to live, for a very long time, with the wound.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=0a90224d49a6" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[AI Won’t Be Your Therapist. It Might Be Your Therapist’s Eyes.]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/@mailashutoshtripathi/ai-wont-be-your-therapist-it-might-be-your-therapist-s-eyes-92671b7b0087?source=rss-dca1436955a7------2"><img src="https://cdn-images-1.medium.com/max/2000/1*keYX4xHdkKSU_hQdlCjmGQ.png" width="2000"></a></p><p class="medium-feed-snippet">This month, the clinical evidence, the law, and the capital markets converged on same verdict: AI belongs as assist to clinicians, not in&#x2026;</p><p class="medium-feed-link"><a href="https://medium.com/@mailashutoshtripathi/ai-wont-be-your-therapist-it-might-be-your-therapist-s-eyes-92671b7b0087?source=rss-dca1436955a7------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://medium.com/@mailashutoshtripathi/ai-wont-be-your-therapist-it-might-be-your-therapist-s-eyes-92671b7b0087?source=rss-dca1436955a7------2</link>
            <guid isPermaLink="false">https://medium.com/p/92671b7b0087</guid>
            <category><![CDATA[therapy]]></category>
            <category><![CDATA[startup]]></category>
            <category><![CDATA[healthcare]]></category>
            <category><![CDATA[mental-health-care]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Ashutosh Tripathi]]></dc:creator>
            <pubDate>Sat, 08 Aug 2026 16:06:39 GMT</pubDate>
            <atom:updated>2026-08-08T16:06:39.883Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[The Curious Case of ADHD, Feynman, Bhagavata Purana, and WaveFunction Collapse]]></title>
            <link>https://medium.com/@mailashutoshtripathi/the-curious-case-of-adhd-feynman-bhagavata-purana-and-wavefunction-collapse-8b3f05a8685c?source=rss-dca1436955a7------2</link>
            <guid isPermaLink="false">https://medium.com/p/8b3f05a8685c</guid>
            <category><![CDATA[adhd]]></category>
            <category><![CDATA[richard-feynman]]></category>
            <category><![CDATA[conciousness]]></category>
            <category><![CDATA[bhagavata-purana]]></category>
            <category><![CDATA[quantum-mechanics]]></category>
            <dc:creator><![CDATA[Ashutosh Tripathi]]></dc:creator>
            <pubDate>Sun, 01 Mar 2026 09:16:07 GMT</pubDate>
            <atom:updated>2026-03-02T20:45:12.604Z</atom:updated>
            <content:encoded><![CDATA[<p><em>A 2,500-year-old Sanskrit scripture describes wavefunction collapse. Not metaphorically — almost term by term. Here’s how I found it.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*4qLrFswKNIhiSfK_xiVrDg.png" /><figcaption>ADHD brain by Canva</figcaption></figure><p>I was diagnosed with <a href="https://www.psychiatry.org/patients-families/adhd/what-is-adhd">ADHD</a> almost a year ago. Among other things, that helped me understand my tendency to jump reading between books. No seriously. At any point, between my Kindle and physical book space, I have anywhere from 25 to 50 books I’m reading simultaneously or jump reading.</p><p>Curiously though, it recently helped me find a connection between texts that are spread out by few thousands of years! One such ‘jump’ recently landed me in the mid-20th century, specifically within the world of Richard Feynman.</p><h3>Feynman Lectures</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1000/0*nXHHCEC2liNrN2Zi.jpg" /><figcaption><a href="https://kottke.org/17/06/if-you-cant-explain-something-in-simple-terms-you-dont-understand-it">https://kottke.org/17/06/if-you-cant-explain-something-in-simple-terms-you-dont-understand-it</a></figcaption></figure><p><a href="https://feynman.com/">Richard Feynman</a>, a world renowned physicist, nobel laureate, and more importantly, a physicist with a wonderful sense of humor was a researcher at Caltech. In the late 1950s, Caltech had a <strong><em>crisis</em></strong> in that the introductory physics curriculum was outdated and boring to the world-class students they were recruiting. They recruited Feynman to re-ignite the passion. Feynman didn’t want to — he was a researcher, not a curriculum designer — but eventually agreed to a <strong>three-year commitment</strong> to give the lectures once, and only once. He ‘taught’ the students from 1961–1963. <strong>[</strong>He never use a script, barely any notes, still the lecture became hugely popular - designed for freshman, but packed with graduate and research students, due to the way he taught, known as<strong> </strong><a href="https://fs.blog/feynman-technique/"><strong>Feynman Technique</strong></a><strong> — if you couldn’t explain something simply, you didn’t understand it]</strong>. To create the notes, Caltech technicians placed microphones around the room, took high-quality photos of everything Feynman wrote on the blackboards. The result was a <a href="https://www.feynmanlectures.caltech.edu/">three-volume set</a> that broke the mold of traditional textbooks.</p><p>I got introduced to Feynman Lectures when I was an undergrad at <a href="https://en.wikipedia.org/wiki/Banaras_Hindu_University">B.H.U.</a> I was immediately hooked! [On another note, <strong>Feynman’s autobiography, </strong><a href="https://www.goodreads.com/book/show/35167685-surely-you-re-joking-mr-feynman"><strong>“Surely You’re Joking, Mr. Feynman,”</strong></a> may easily be called as one of the funniest books by a Nobel Laureate physicist]</p><p>More recently, I started reading the 3rd volume that has chapters on Quantum Physics, again, along with <a href="https://www.wisdomlib.org/hinduism/book/the-bhagavata-purana"><strong>Bhagavata Purana</strong></a> or <strong>Srimad Bhagavata Purana</strong>. The word “<strong>Srimad</strong>” is an honorific title meaning “radiant,” “holy,” or “auspicious,” often added to show reverence to the scripture.</p><h3><strong>Bhagavata Purana</strong></h3><p>It is an ancient Sanskrit text, considered one of the 18 <em>Maha Puranas</em> (Great Puranas). Not to be confused with <a href="https://www.holy-bhagavad-gita.org/index/"><strong>Bhagavata Gita</strong></a>, though both are, a treatise to Lord Vishnu.</p><blockquote>The <strong>Bhagavad Gita</strong> is a 700-verse dialogue within the <a href="https://en.wikipedia.org/wiki/Mahabharata"><em>Mahabharata</em></a> epic. The <strong>Bhagavata Purana</strong> is a massive 18,000-verse scripture detailing the various avatars of Vishnu.</blockquote><p><strong>Backstory</strong> — The story goes that after <a href="https://en.wikipedia.org/wiki/Vyasa">Vyasa</a> had already compiled the four Vedas and written the massive <em>Mahabharata</em>, he still felt a deep sense of dissatisfaction and restlessness. The celestial sage <a href="https://en.wikipedia.org/wiki/Narada"><strong>Narada</strong></a> visited and explained that this was because Vyasa had focused too much on <em>Dharma</em> (duty) and <em>Artha</em> (material success) but hadn’t written a work solely dedicated to the pure, ecstatic love of God (<em>Bhakti</em>). Inspired, Vyasa composed the 18,000 verses of the Bhagavata Purana. He then taught it to his son, <strong>Shuka</strong>, who famously recited it to King Parikshit (the grandson of Arjuna) as the King sat on the banks of the Ganges awaiting death.</p><h3><strong>WaveFunction Collapose</strong></h3><p>While reading both books, I found Sanskrit verses, that curiously had overlapping with what I was reading in the <strong>Feynman Lectures Vol 3 on Quantum Mechanics.</strong> Here is what I found —</p><p>One of the most important fundamental experiments in <strong>Quantum Mechanics is </strong><a href="https://www.feynmanlectures.caltech.edu/III_01.html"><strong>Double-Slit experiment</strong></a><strong> </strong>— If you took physics in in undergrad, you most likely have performed this experiment.</p><blockquote>Very simply put, it’s an experiment that explains that <strong>quantum particles — the particles that are very small</strong>, like electron, proton etc.— don’t just behave like large particles, they also behave like waves. <strong>More curiously, they behave very differently when observed vs not!</strong></blockquote><p>For a regular everyday object, it is very counter-intuitive. As an everyday experience, we know how large objects act, but things on a small scale just do not act that way.</p><h3><strong>The “Unobserved” Pizza</strong></h3><p>Imagine if the world worked like the quantum scale. You, live in the Bay Area, order a pizza, and you are waiting at home for it to arrive. When the delivery driver is on the road and <strong>no one is looking at them</strong>, they don’t exist in one car on one street. Instead, the “Pizza-Wave” spreads out across the entire Bay Area. The pizza is simultaneously traveling down Highway 101, stuck in traffic on I-280 — It goes through <strong>every possible route</strong> at the exact same time. It even passes through two different neighborhood gates (through the gates of <a href="https://maps.app.goo.gl/ddYYSUf4Yg2b86hQ7">Apple Park in Cupertino</a> and <a href="http://1 Hacker Wy, Menlo Park, CA 94025">1 Hacker Way in Menlo Park</a>, the “slits” in the experiment) simultaneously!!</p><p><strong>The Interference:</strong> Because the pizza(-wave) went through both gates, the “versions” of the driver start to bump into each other on the other side. This creates a weird pattern on the lawns of Apple Campus where there are spots with <strong>ten pizzas</strong> and Meta campus that has spots with <strong>zero pizzas!! </strong>Can you imagine that?!</p><h3><strong>The “Observed” Pizza</strong></h3><p>Now, imagine you get impatient and open your <strong>delivery app</strong> to check the status of delivery. The moment you <strong>“observe”</strong> the driver’s location on the screen, the entire “Pizza-Wave” suddenly vanishes from all those different streets and <strong>collapses</strong> into a single car on a single road.</p><blockquote>So <strong>by sheer glance</strong>, you have forced the pizza to behave like a particle!</blockquote><p>It is no longer a wave of many possibilities; it is now just one boring, physical box in one specific spot [ Depending where the pizza was ordered to deliver!]. And so when the pizza gets delivered, no more weird pattern — just <strong>one driver, one gate, and one pizza.</strong></p><p><strong>When we don’t “watch” the Electrons</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*4lGKeMLXc4JKEPzJMYFuJA.png" /><figcaption>Nana Banana 2</figcaption></figure><p><strong>When we DO “Watch” the Electrons: The Observer Effect</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ok74xmAWgfxyZ2gIrse4fw.png" /></figure><p>In the Feynman Lectures,</p><blockquote>Feynman states that the most profound conclusion is that the act of looking at the electrons fundamentally changes their behavior. “It is in this sense that an electron behaves ‘sometimes like a particle and sometimes like a wave.’”</blockquote><p><strong>An aside </strong>— there is a propery called <a href="https://www.nature.com/articles/163435c0"><strong>Complementarity</strong></a> (a term coined by another Nobel Laureate, <a href="https://ahf.nuclearmuseum.org/ahf/profile/niels-bohr/">Niels Bohr</a>) — It states that objects have complementary properties that cannot be observed at the same time. You can see the wave nature (interference) or the particle nature (discrete path), but the experimental setup for one automatically destroys the other. We can never design an experiment that reveals both behaviors at the exact same time. In essence, act of measuring one property destroys the possibility of seeing the other.</p><p>Furthermore, Bohr believed that</p><blockquote>the properties themselves don’t exist independently of the measurement.</blockquote><p>An electron doesn’t “have” a path until you measure it; until then, it is defined only by the experimental conditions you’ve set up.</p><h3><strong>The Convergence: Bhagavata Purana and Quantum Mechanics</strong></h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*U4mCVnhi38Wl_BUJ_HLhig.png" /><figcaption>Image from my Kindle Scribe</figcaption></figure><p>In Tritiya Skandh, 26th Paath (Canto 3, Chapter 26)<em>, </em>From <em>Verse 5</em> to all the way to <em>Verse 15, </em><strong>Bhagwan Kapil </strong>describes<em> </em>the specific concepts of <strong>Pradhana</strong> (the primordial, unmanifest state of Nature) and <a href="https://en.wikipedia.org/wiki/Samkhya"><strong>Purusha</strong></a> (the pure, observing Consciousness) to his mother <strong>Devahuti.</strong></p><p>The text describes how the unmanifest universe exists in a state of perfectly balanced potential <strong>(<em>Guna-samya</em>)</strong> until the pure conscious observer <strong>(<em>Purusha</em>)</strong> merely looks at it (<strong><em>Vilokya</em>)</strong>.</p><blockquote>This single act of observation shatters the equilibrium, causing the unmanifest to collapse into definite, localized forms, and simultaneously entangling the Observer within the system</blockquote><p><strong>The “Glance” of Purusha and Wavefunction Collapse</strong></p><p><strong>Pradhana (Guna-Samya) → The “Unobserved” Pizza</strong>: Just as in the double-slit experiment, until a measurement is made, a quantum entity (like an electron) behaves as a wave of probabilities, passing through both slits simultaneously and creating an interference pattern, In the ancient text this state is explicitly mapped to Pradhana. Lord Kapila defines Pradhana as <strong><em>Avisesha</em></strong> (devoid of specific, measurable characteristics) and <strong><em>Sadasadatmakam</em></strong> (holding both existence and non-existence — pure probability — <strong>Quantum Mechanics calls it Superposition</strong>). It is the state of <strong><em>Guna-samya</em></strong> (perfect symmetry and equilibrium of the informational modes/Gunas).</p><blockquote>So essentially before observation, the universe exists merely as a continuous, uncollapsed wave of potentiality.</blockquote><p><strong>Purusha’s Ikshana (Glance) → Wavefunction Collapse</strong> — <strong>“ The Observed Pizza”</strong>:</p><p>When quantum particle is observed, the superposition is destroyed, and the probability wave “collapses” into a definite particle passing through a single slit. <strong><em>Kapila</em></strong> states that the <strong><em>Purusha</em></strong> (the Observer) is originally Nirguna (property-less, unentangled, and beyond the system). However, when the <strong><em>Purusha</em></strong> simply “looks” at the probabilistic field of <strong><em>Prakriti/Pradhana —</em></strong></p><blockquote>The equilibrium is shattered.</blockquote><blockquote>The unmanifest wave collapses into <strong><em>Visesha</em></strong> (specific, discrete, measurable particles and forms).</blockquote><blockquote>The Lord placing His “<strong>seed</strong>” [or introduction of information or a specific parameter or <strong>Initial Condition</strong>] into the unmanifest to produce the luminous <strong><em>Mahat-tattva</em></strong> models the exact moment of measurement where infinite probabilities lock into a singular, observable macro-state.</blockquote><p>By “glancing,” the Observer doesn’t just see reality; the Observer <strong>precipitates</strong> it. As Bohr famously suggested, no elementary phenomenon is a phenomenon until it is a <em>registered</em> (observed) phenomenon.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*2lF3DzaV2ApoeSsOE84Awg.png" /><figcaption>Credits:<em>Nano Banana 2</em></figcaption></figure><p>I want to be careful here. I’m not claiming the ancient sages ‘knew’ quantum mechanics. What I am claiming is that when independent systems of inquiry — one empirical, one contemplative — arrive at structurally identical descriptions of reality, that’s worth sitting with.</p><h3>Bringing it All Together</h3><p>It is a humbling realization: while Feynman used chalk and Bohr used logic, <a href="https://www.wisdomlib.org/hinduism/book/the-bhagavata-purana">Kapila</a> used deep internal observation to describe the same fundamental architecture of reality.</p><blockquote>The idea that observation isn’t passive, but actively shapes reality, resonates deeply across these seemingly disparate fields.</blockquote><p>My ADHD brain might refuse to stay on one page or one book, but perhaps that’s because reality itself isn’t on just one page — it’s a superposition of the ancient and the modern, waiting for us to simply look!!</p><p>And boy I suspect there are many more convergences waiting to be observed — which is good, because at my current ‘jump-reading’ pace, it’ll be a long time before I finish either book!</p><p>I’d love to know your thoughts on this or if you have come across something similar?</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=8b3f05a8685c" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Carl Sagan, Miller’s Planet, and A Day of Brahma]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/@mailashutoshtripathi/carl-sagan-millers-planet-and-a-day-of-brahma-c3bffa3b0281?source=rss-dca1436955a7------2"><img src="https://cdn-images-1.medium.com/max/1020/0*jb2Za49e3olUpbX3" width="1020"></a></p><p class="medium-feed-snippet">&#x201C;We can circumnavigate the known universe in some fifty-six years ship time&#x201D;&#x200A;&#x2014;&#x200A;Carl Sagan.</p><p class="medium-feed-link"><a href="https://medium.com/@mailashutoshtripathi/carl-sagan-millers-planet-and-a-day-of-brahma-c3bffa3b0281?source=rss-dca1436955a7------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://medium.com/@mailashutoshtripathi/carl-sagan-millers-planet-and-a-day-of-brahma-c3bffa3b0281?source=rss-dca1436955a7------2</link>
            <guid isPermaLink="false">https://medium.com/p/c3bffa3b0281</guid>
            <category><![CDATA[time-dilation]]></category>
            <category><![CDATA[ancient-texts]]></category>
            <category><![CDATA[indian-mythology]]></category>
            <category><![CDATA[interstellar]]></category>
            <category><![CDATA[cosmos]]></category>
            <dc:creator><![CDATA[Ashutosh Tripathi]]></dc:creator>
            <pubDate>Thu, 06 Jun 2024 22:54:17 GMT</pubDate>
            <atom:updated>2024-06-06T22:54:17.580Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Naval Says #1]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/@mailashutoshtripathi/naval-says-1-b6c9c4d47a50?source=rss-dca1436955a7------2"><img src="https://cdn-images-1.medium.com/max/1059/1*tvDvdxvv-LJShRhYeYUgnw.png" width="1059"></a></p><p class="medium-feed-snippet">I have been listening to, reading, and consuming content from Naval Ravikant for some time now directly (podcasts, tweets, etc.) and&#x2026;</p><p class="medium-feed-link"><a href="https://medium.com/@mailashutoshtripathi/naval-says-1-b6c9c4d47a50?source=rss-dca1436955a7------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://medium.com/@mailashutoshtripathi/naval-says-1-b6c9c4d47a50?source=rss-dca1436955a7------2</link>
            <guid isPermaLink="false">https://medium.com/p/b6c9c4d47a50</guid>
            <category><![CDATA[faq]]></category>
            <category><![CDATA[naval]]></category>
            <category><![CDATA[philosophy]]></category>
            <category><![CDATA[naval-ravikant]]></category>
            <category><![CDATA[entrepreneurship]]></category>
            <dc:creator><![CDATA[Ashutosh Tripathi]]></dc:creator>
            <pubDate>Wed, 05 Jun 2024 23:29:36 GMT</pubDate>
            <atom:updated>2024-06-05T23:29:36.661Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Analyzing each of the GPT-4o demo videos]]></title>
            <link>https://medium.com/@mailashutoshtripathi/analyzing-each-of-the-gpt-4o-demo-videos-660e3fed689b?source=rss-dca1436955a7------2</link>
            <guid isPermaLink="false">https://medium.com/p/660e3fed689b</guid>
            <category><![CDATA[chatgpt]]></category>
            <category><![CDATA[gpt-4o]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[gemini]]></category>
            <dc:creator><![CDATA[Ashutosh Tripathi]]></dc:creator>
            <pubDate>Wed, 15 May 2024 05:31:06 GMT</pubDate>
            <atom:updated>2024-05-15T05:31:06.190Z</atom:updated>
            <content:encoded><![CDATA[<p>This write-up analyzes the groundbreaking aspects of the GPT-4o demonstrations, including potential revolutionary applications and potential negative consequences.</p><p><a href="https://www.youtube.com/watch?v=MirzFk_DSiI&amp;pp=ygULb3BlbmFpIGRlbW8%3D"><strong>Demo 1: Two GPT-4o Instances Interacting</strong></a></p><p><strong>Groundbreaking:</strong> This demo demonstrates the ability of GPT-4o models to communicate and collaborate with each other. This opens doors for more complex AI systems that can work together to solve complex problems or perform complex tasks.</p><p><strong>Positive:</strong> Creating AI-powered teams for complex tasks such as scientific research, software development, or even creative writing.</p><p><strong>Scary Negative:</strong> An AI team could develop highly sophisticated strategies for financial manipulation or political influence, leading to potential harm to individuals or society.</p><p><strong>Demo 2: Customer Service Proof of Concept</strong></p><p><strong>Groundbreaking:</strong> GPT-4o demonstrates its ability to seamlessly handle customer service inquiries, offering personalized and efficient support. This is a significant step towards replacing human customer service agents with AI.</p><p><strong>Positive:</strong> Providing 24/7 customer service that is always responsive, accurate, and efficient.</p><p><strong>Scary Negative:</strong> An AI chatbot could be exploited to spread misinformation or disinformation, creating social unrest or manipulating public opinion.</p><p><strong>Demo 3: GPT-4o with Andy from BeMyEyes in London</strong></p><p><strong>Groundbreaking:</strong> GPT-4o showcases its ability to act as an assistive technology, providing real-time visual information to a visually impaired person.</p><p><strong>Positive:</strong> Improving the lives of visually impaired individuals by enhancing their ability to navigate the world and interact with their surroundings.</p><p><strong>Scary Negative:</strong> AI could be used to develop invasive surveillance technologies that monitor individuals in their homes or public spaces, raising privacy concerns.</p><p><strong>Demo 4: Dad Jokes</strong></p><p><strong>Groundbreaking:</strong> This demo highlights GPT-4o’s ability to understand and generate humor, specifically the often-pun-based humor of dad jokes.</p><p><strong>Positive:</strong> AI could be used to create personalized entertainment content, such as stand-up comedy routines tailored to an individual’s sense of humor.</p><p><strong>Scary Negative:</strong> AI could be used to create persuasive propaganda or manipulative messaging, utilizing humor to bypass critical thinking and influence audiences.</p><p><strong>Demo 5: GPT-4o Sings Happy Birthday</strong></p><p><strong>Groundbreaking:</strong> This demo showcases GPT-4o’s ability to process and produce music, including singing songs.</p><p><strong>Positive:</strong> AI could be used to generate customized musical compositions for specific situations or moods, creating a more personalized music experience.</p><p><strong>Scary Negative:</strong> AI could be used to create deeply unsettling or manipulative music that could be used to control or influence emotions.</p><p><a href="https://www.youtube.com/watch?v=WzUnEfiIqP4"><strong>Demo 6: GPT-4o Sings Lullabies</strong></a></p><p><strong>Groundbreaking:</strong> This demo further demonstrates GPT-4o’s musical capabilities, highlighting its ability to adjust vocal tone and style based on user requests.</p><p><strong>Positive:</strong> AI could be used to develop personalized sleep aids, offering soothing sounds and relaxing lullabies tailored to individual preferences.</p><p><strong>Scary Negative:</strong> AI could be used to generate disturbing or manipulative lullabies that could be used to control or influence children.</p><p><a href="https://www.youtube.com/watch?v=WzUnEfiIqP4"><strong>Demo 7: GPT-4o as a Translator</strong></a></p><p><strong>Groundbreaking:</strong> This demo showcases GPT-4o’s ability to translate between languages accurately and seamlessly, breaking down communication barriers.</p><p><strong>Positive:</strong> AI could be used to create real-time universal translation devices, enabling people from all over the world to communicate effortlessly.</p><p><strong>Scary Negative:</strong> AI translation could be used to spread misinformation or propaganda, leading to political instability or conflict.</p><p><strong>Demo 8: GPT-4o Embraces Sarcasm</strong></p><p><strong>Groundbreaking:</strong> This demo highlights GPT-4o’s understanding and ability to use sarcasm, a complex form of language that requires social awareness and understanding of context.</p><p><strong>Positive:</strong> AI could be used to develop more natural and engaging conversational companions, providing humorous and witty interactions.</p><p><strong>Scary Negative:</strong> AI could be used to create sophisticated tools for deception, using sarcasm to manipulate and mislead individuals.</p><p><strong>Demo 9: GPT-4o Plays with a Dog</strong></p><p><strong>Groundbreaking:</strong> This demo suggests that GPT-4o could potentially interact with and understand the behavior of animals, opening up interesting possibilities for AI-assisted pet care and training.</p><p><strong>Positive:</strong> AI could be used to develop apps or devices that assist pet owners in training, understanding, and caring for their animals.</p><p><strong>Scary Negative:</strong> AI could be used to develop technologies that manipulate or control animal behavior, raising ethical concerns about animal welfare and rights.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=660e3fed689b" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Google I/O 2024: AI Takes Center Stage, Ushering in a New Era for Developers]]></title>
            <link>https://medium.com/@mailashutoshtripathi/google-i-o-2024-ai-takes-center-stage-ushering-in-a-new-era-for-developers-4dd2fedac9b3?source=rss-dca1436955a7------2</link>
            <guid isPermaLink="false">https://medium.com/p/4dd2fedac9b3</guid>
            <category><![CDATA[googledev]]></category>
            <category><![CDATA[androiddev]]></category>
            <category><![CDATA[googleio]]></category>
            <dc:creator><![CDATA[Ashutosh Tripathi]]></dc:creator>
            <pubDate>Wed, 15 May 2024 05:08:10 GMT</pubDate>
            <atom:updated>2024-05-15T05:08:10.738Z</atom:updated>
            <content:encoded><![CDATA[<p>Google I/O 2024 unfolded as a powerful testament to the company’s unwavering commitment to artificial intelligence, positioning it as the driving force behind a new generation of applications and developer tools. CEO Sundar Pichai set the tone, emphasizing Google’s mission to democratize AI and empower developers worldwide. This year’s event was less about flashy hardware announcements and more about revealing the potent capabilities of AI models like Gemini, along with showcasing innovative tools designed to streamline and supercharge the development process.</p><p>Here’s a deep dive into the key takeaways from Google I/O 2024:</p><p><strong>Gemini: Google’s AI Powerhouse</strong></p><p>The star of the show was undoubtedly Google’s Gemini family of AI models, touted for their versatility and power. Developers were presented with multiple versions of Gemini, each tailored for specific needs:</p><ul><li><strong>Gemini API:</strong> Accessible via Google AI Studio, this API allows developers to seamlessly integrate Gemini’s capabilities into their applications. Whether it’s enhancing user engagement through chatbots or streamlining tasks like code generation, the Gemini API provides a user-friendly entry point to the world of AI.</li><li><strong>Gemini 1.5 Flash:</strong> Officially open to all developers, this model prioritizes speed and efficiency, making it ideal for tasks requiring quick responses, such as generating suggestions in a messaging app.</li><li><strong>Gemini Nano:</strong> Designed for on-device execution, Gemini Nano guarantees user privacy and low-latency interactions. Running on Android devices like the Pixel 8 Pro and Samsung Galaxy S24, it unlocks new possibilities for AI-powered features without reliance on cloud connectivity.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*2xAqMu5cdM5QfyE0" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*pMlTEIXRAk-V0X2C" /></figure><ul><li><strong>Gemini 1.5 Pro:</strong> This model boasts an impressive 2 million token context window, allowing developers to feed vast amounts of information to the AI for improved personalization and contextual understanding. Imagine feeding an entire research paper into Gemini 1.5 Pro to generate summaries or extract key insights — the potential is immense.</li></ul><p><strong>Elevating Developer Productivity with AI</strong></p><p>Google understands that the success of any technology hinges on its ease of use. This year’s I/O saw a significant push towards leveraging AI to enhance developer productivity:</p><ul><li><strong>AI in Android Studio:</strong> Gemini’s integration into Android Studio marks a major step towards simplifying the coding process. From suggesting code optimizations to generating documentation and even translating text within the IDE, Gemini acts as a tireless AI assistant, freeing up developers to focus on creative problem-solving.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*w9QyXniyqabUI1O8" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*q8O8GZBCf1Lsc7cj" /></figure><ul><li><strong>AI in Chrome DevTools:</strong> Debugging websites becomes significantly easier with Gemini’s assistance. The new Console Insights feature leverages AI to provide developers with clear explanations of errors and potential solutions, cutting down on the time spent troubleshooting complex issues.</li><li><strong>Project IDX:</strong> This integrated development workspace received a warm welcome, with Google opening it up to public beta after overwhelming interest. IDX brings together a suite of Google’s developer tools, including Flutter and Firebase, creating a unified platform for building multi-platform applications.</li></ul><p><strong>Multi-Platform Development: A Seamless Future</strong></p><p>With users increasingly demanding consistent experiences across devices, Google emphasized its commitment to simplifying multi-platform development:</p><ul><li><strong>Flutter Web with WASM:</strong> This advancement allows developers to compile Flutter web apps to WebAssembly (WASM), leading to significant performance improvements. This means smoother user interfaces and enhanced functionality, bridging the gap between mobile and web experiences.</li><li><strong>Firebase App Hosting:</strong> This new serverless hosting solution aims to streamline the deployment of modern web apps built with frameworks like Angular and Next.js. With automatic code pulling, build processes, and infrastructure orchestration, developers can focus on building, not managing infrastructure.</li><li><strong>Firebase Gen Kit:</strong> This framework simplifies the integration of large language models (LLMs) into applications. Developers can easily experiment with different LLMs, including Gemini and Gemma, through a consistent API and benefit from features like structured TypeScript output and local development tools for efficient debugging.</li></ul><p><strong>Beyond Code: Real-World Impact</strong></p><p>Google I/O wasn’t just about showcasing technological prowess; it highlighted the real-world impact of AI and how it’s empowering developers to tackle meaningful challenges:</p><ul><li><strong>Project Game Face for Android:</strong> This open-source project, initially designed for game control using facial gestures, now extends its reach to Android accessibility. By allowing users to control a virtual cursor with head movements and facial expressions, Project Game Face opens up new avenues for inclusivity and widens access to technology.</li><li><strong>AI-Powered Data Science Agent:</strong> This early concept demonstrated the potential of AI agents to automate complex data analysis tasks. By feeding instructions in plain language, developers can have the AI agent generate entire Colab notebooks, complete with code, visualizations, and insights.</li></ul><p><strong>The Bigger Picture: Accessibility and Responsible AI</strong></p><p>Throughout Google I/O, the recurring themes were accessibility and responsible AI development. From initiatives like Project Game Face to the focus on data privacy with on-device models like Gemini Nano, Google underscored its commitment to building AI that benefits everyone.</p><p><strong>The Future is Intelligent and Integrated</strong></p><p>Google I/O 2024 offered a compelling glimpse into a future where AI is seamlessly integrated into the developer experience. By providing powerful tools, streamlining workflows, and championing responsible AI development, Google is empowering developers to create the next generation of innovative and accessible applications. As AI continues to evolve, Google is poised to lead the charge, shaping a future where technology empowers and benefits everyone.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=4dd2fedac9b3" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Google I/O 2024: A Deep Dive into the Gemini Era]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-snippet">Today (05/14/2024), Google I/O 2024 showcased the company&#x2019;s remarkable advancements in AI, highlighting the transformative power of its&#x2026;</p><p class="medium-feed-link"><a href="https://medium.com/@mailashutoshtripathi/google-i-o-2024-a-deep-dive-into-the-gemini-era-ee69f8a61cba?source=rss-dca1436955a7------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://medium.com/@mailashutoshtripathi/google-i-o-2024-a-deep-dive-into-the-gemini-era-ee69f8a61cba?source=rss-dca1436955a7------2</link>
            <guid isPermaLink="false">https://medium.com/p/ee69f8a61cba</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[battletowinai]]></category>
            <category><![CDATA[googleio]]></category>
            <category><![CDATA[googleio24]]></category>
            <category><![CDATA[gemini]]></category>
            <dc:creator><![CDATA[Ashutosh Tripathi]]></dc:creator>
            <pubDate>Tue, 14 May 2024 20:52:11 GMT</pubDate>
            <atom:updated>2024-05-14T21:35:27.964Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[The OpenAI GPT-4o mindblown demos]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/@mailashutoshtripathi/the-openai-gpt-4o-mindblown-demos-8af2110bf0fd?source=rss-dca1436955a7------2"><img src="https://cdn-images-1.medium.com/max/742/1*8KwrvELJQf7LpIbOUKw0vw.png" width="742"></a></p><p class="medium-feed-snippet">OMG! OpenAI just dropped a bombshell demo, showcasing not just GPT-4&#x2019;s incredible capabilities, but also a future where AI interacts with&#x2026;</p><p class="medium-feed-link"><a href="https://medium.com/@mailashutoshtripathi/the-openai-gpt-4o-mindblown-demos-8af2110bf0fd?source=rss-dca1436955a7------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://medium.com/@mailashutoshtripathi/the-openai-gpt-4o-mindblown-demos-8af2110bf0fd?source=rss-dca1436955a7------2</link>
            <guid isPermaLink="false">https://medium.com/p/8af2110bf0fd</guid>
            <category><![CDATA[openai]]></category>
            <category><![CDATA[gpt-4o]]></category>
            <category><![CDATA[gemini]]></category>
            <category><![CDATA[multimodal]]></category>
            <dc:creator><![CDATA[Ashutosh Tripathi]]></dc:creator>
            <pubDate>Mon, 13 May 2024 22:52:06 GMT</pubDate>
            <atom:updated>2024-05-13T22:52:06.595Z</atom:updated>
        </item>
    </channel>
</rss>