<?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 Armin Norouzi, Ph.D on Medium]]></title>
        <description><![CDATA[Stories by Armin Norouzi, Ph.D on Medium]]></description>
        <link>https://medium.com/@arminnorouzi?source=rss-7d14ec5b901a------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*71_WV694bBLRLPlXkAgRCQ.png</url>
            <title>Stories by Armin Norouzi, Ph.D on Medium</title>
            <link>https://medium.com/@arminnorouzi?source=rss-7d14ec5b901a------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Tue, 17 Nov 2026 17:45:47 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@arminnorouzi/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[AI Reasoning vs. Human Thinking: What’s the Difference?]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/data-science-collective/ai-reasoning-vs-human-thinking-whats-the-difference-e343e7a2d874?source=rss-7d14ec5b901a------2"><img src="https://cdn-images-1.medium.com/max/2400/1*-iCFO3Sqpxl67jRqpx5PKQ.jpeg" width="2400"></a></p><p class="medium-feed-snippet">Does AI truly understand? See how LLMs reason on GPUs (KV cache, FlashAttention, GRPO) and how that compares to the neuroscience of the&#x2026;</p><p class="medium-feed-link"><a href="https://medium.com/data-science-collective/ai-reasoning-vs-human-thinking-whats-the-difference-e343e7a2d874?source=rss-7d14ec5b901a------2">Continue reading on Data Science Collective »</a></p></div>]]></description>
            <link>https://medium.com/data-science-collective/ai-reasoning-vs-human-thinking-whats-the-difference-e343e7a2d874?source=rss-7d14ec5b901a------2</link>
            <guid isPermaLink="false">https://medium.com/p/e343e7a2d874</guid>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[deep-learning]]></category>
            <category><![CDATA[llm]]></category>
            <category><![CDATA[neuroscience]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Armin Norouzi, Ph.D]]></dc:creator>
            <pubDate>Wed, 11 Nov 2026 00:33:56 GMT</pubDate>
            <atom:updated>2026-11-14T15:18:10.950Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[The Engineering Proof of the 4% Rule: Lyapunov Stability for Retirement Withdrawals]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.datadriveninvestor.com/the-engineering-proof-of-the-4-rule-lyapunov-stability-for-retirement-withdrawals-0207feef8fda?source=rss-7d14ec5b901a------2"><img src="https://cdn-images-1.medium.com/max/1650/0*A5fgKm2qiOioBUTL.png" width="1650"></a></p><p class="medium-feed-snippet">The 4% rule is the most important constant in retirement planning, and almost nobody checks it the way an engineer checks a constant.</p><p class="medium-feed-link"><a href="https://medium.datadriveninvestor.com/the-engineering-proof-of-the-4-rule-lyapunov-stability-for-retirement-withdrawals-0207feef8fda?source=rss-7d14ec5b901a------2">Continue reading on DataDrivenInvestor »</a></p></div>]]></description>
            <link>https://medium.datadriveninvestor.com/the-engineering-proof-of-the-4-rule-lyapunov-stability-for-retirement-withdrawals-0207feef8fda?source=rss-7d14ec5b901a------2</link>
            <guid isPermaLink="false">https://medium.com/p/0207feef8fda</guid>
            <category><![CDATA[personal-finance]]></category>
            <category><![CDATA[investing]]></category>
            <category><![CDATA[data-driven-finance]]></category>
            <category><![CDATA[retirement]]></category>
            <category><![CDATA[finance]]></category>
            <dc:creator><![CDATA[Armin Norouzi, Ph.D]]></dc:creator>
            <pubDate>Mon, 19 Oct 2026 05:17:58 GMT</pubDate>
            <atom:updated>2026-10-27T09:58:55.423Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[The Evaluation Stack: Metrics That Predict Production Quality]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://pub.towardsai.net/the-evaluation-stack-metrics-that-predict-production-quality-46e37ba427fd?source=rss-7d14ec5b901a------2"><img src="https://cdn-images-1.medium.com/max/989/1*tUnNJmN5HyaFpOFLX-pCiA.png" width="989"></a></p><p class="medium-feed-snippet">The hardest engineering problem in LLM deployment is not latency or cost &#x2014; it is knowing whether your model got better. Metrics that seem&#x2026;</p><p class="medium-feed-link"><a href="https://pub.towardsai.net/the-evaluation-stack-metrics-that-predict-production-quality-46e37ba427fd?source=rss-7d14ec5b901a------2">Continue reading on Towards AI »</a></p></div>]]></description>
            <link>https://pub.towardsai.net/the-evaluation-stack-metrics-that-predict-production-quality-46e37ba427fd?source=rss-7d14ec5b901a------2</link>
            <guid isPermaLink="false">https://medium.com/p/46e37ba427fd</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[test-autoevaluacion]]></category>
            <category><![CDATA[llm-automation]]></category>
            <category><![CDATA[llm-evaluation]]></category>
            <dc:creator><![CDATA[Armin Norouzi, Ph.D]]></dc:creator>
            <pubDate>Mon, 24 Aug 2026 00:01:02 GMT</pubDate>
            <atom:updated>2026-08-24T00:01:02.628Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Tool Call Orchestration: Sequential, Parallel, and DAG Execution]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://pub.towardsai.net/tool-call-orchestration-sequential-parallel-and-dag-execution-53a73b4b076d?source=rss-7d14ec5b901a------2"><img src="https://cdn-images-1.medium.com/max/996/1*s8UcQmYcNGjBj5kC_lA7qw.png" width="996"></a></p><p class="medium-feed-snippet">A research agent calling 6 tools sequentially waits 1,832ms at P50. The same 6 tools run in parallel &#x2014; ignoring all dependencies &#x2014; finish&#x2026;</p><p class="medium-feed-link"><a href="https://pub.towardsai.net/tool-call-orchestration-sequential-parallel-and-dag-execution-53a73b4b076d?source=rss-7d14ec5b901a------2">Continue reading on Towards AI »</a></p></div>]]></description>
            <link>https://pub.towardsai.net/tool-call-orchestration-sequential-parallel-and-dag-execution-53a73b4b076d?source=rss-7d14ec5b901a------2</link>
            <guid isPermaLink="false">https://medium.com/p/53a73b4b076d</guid>
            <category><![CDATA[llm-agent]]></category>
            <category><![CDATA[llm-agent-frameworks]]></category>
            <category><![CDATA[agent-orchestration]]></category>
            <category><![CDATA[ai-agent-orchestration]]></category>
            <dc:creator><![CDATA[Armin Norouzi, Ph.D]]></dc:creator>
            <pubDate>Tue, 18 Aug 2026 12:31:01 GMT</pubDate>
            <atom:updated>2026-08-18T12:31:01.327Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Open Source vs. Closed API LLMs: A 2026 Production Decision Guide]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://levelup.gitconnected.com/open-source-vs-closed-api-llms-a-2026-production-decision-guide-e8dcc7b995a6?source=rss-7d14ec5b901a------2"><img src="https://cdn-images-1.medium.com/max/1000/1*LifZL-fe4Kcm71N7sC0d1g.png" width="1000"></a></p><p class="medium-feed-snippet">The open vs. closed debate in LLMs has been running since GPT-3 launched in 2020, but in 2026 the conversation has fundamentally changed&#x2026;</p><p class="medium-feed-link"><a href="https://levelup.gitconnected.com/open-source-vs-closed-api-llms-a-2026-production-decision-guide-e8dcc7b995a6?source=rss-7d14ec5b901a------2">Continue reading on Level Up Coding »</a></p></div>]]></description>
            <link>https://levelup.gitconnected.com/open-source-vs-closed-api-llms-a-2026-production-decision-guide-e8dcc7b995a6?source=rss-7d14ec5b901a------2</link>
            <guid isPermaLink="false">https://medium.com/p/e8dcc7b995a6</guid>
            <category><![CDATA[open-source]]></category>
            <category><![CDATA[llm]]></category>
            <category><![CDATA[ai-cost-optimization]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[llm-applications]]></category>
            <dc:creator><![CDATA[Armin Norouzi, Ph.D]]></dc:creator>
            <pubDate>Mon, 17 Aug 2026 15:24:13 GMT</pubDate>
            <atom:updated>2026-08-17T15:24:13.989Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Implement Speculative Decoding with a Draft Model in Python]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://pub.towardsai.net/implement-speculative-decoding-with-a-draft-model-in-python-bd2e7e6b483b?source=rss-7d14ec5b901a------2"><img src="https://cdn-images-1.medium.com/max/1169/1*XuSjVweq3CXXV0jG9az1Rw.png" width="1169"></a></p><p class="medium-feed-snippet">Generating text with a large language model is sequential by default: one token at a time, one forward pass per token. On a 70B-parameter&#x2026;</p><p class="medium-feed-link"><a href="https://pub.towardsai.net/implement-speculative-decoding-with-a-draft-model-in-python-bd2e7e6b483b?source=rss-7d14ec5b901a------2">Continue reading on Towards AI »</a></p></div>]]></description>
            <link>https://pub.towardsai.net/implement-speculative-decoding-with-a-draft-model-in-python-bd2e7e6b483b?source=rss-7d14ec5b901a------2</link>
            <guid isPermaLink="false">https://medium.com/p/bd2e7e6b483b</guid>
            <dc:creator><![CDATA[Armin Norouzi, Ph.D]]></dc:creator>
            <pubDate>Sat, 15 Aug 2026 18:01:01 GMT</pubDate>
            <atom:updated>2026-08-15T18:01:01.481Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[The Inference Latency Trap: Why Smaller Isn’t Always Faster]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://levelup.gitconnected.com/the-inference-latency-trap-why-smaller-isnt-always-faster-3f1bc6e1e3e1?source=rss-7d14ec5b901a------2"><img src="https://cdn-images-1.medium.com/max/995/1*cYfs7SF4fJwJ8ECi8klqsQ.png" width="995"></a></p><p class="medium-feed-snippet">There is a popular belief in LLM deployment: &#x201C;use a smaller model if you need lower latency.&#x201D; It is often right. But it is not always&#x2026;</p><p class="medium-feed-link"><a href="https://levelup.gitconnected.com/the-inference-latency-trap-why-smaller-isnt-always-faster-3f1bc6e1e3e1?source=rss-7d14ec5b901a------2">Continue reading on Level Up Coding »</a></p></div>]]></description>
            <link>https://levelup.gitconnected.com/the-inference-latency-trap-why-smaller-isnt-always-faster-3f1bc6e1e3e1?source=rss-7d14ec5b901a------2</link>
            <guid isPermaLink="false">https://medium.com/p/3f1bc6e1e3e1</guid>
            <category><![CDATA[llm]]></category>
            <category><![CDATA[vllm]]></category>
            <category><![CDATA[llm-inference]]></category>
            <category><![CDATA[latency]]></category>
            <category><![CDATA[gpu]]></category>
            <dc:creator><![CDATA[Armin Norouzi, Ph.D]]></dc:creator>
            <pubDate>Tue, 11 Aug 2026 15:49:48 GMT</pubDate>
            <atom:updated>2026-08-11T15:49:48.973Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Introducing Data-Driven Personal Finance]]></title>
            <link>https://arminnorouzi.medium.com/introducing-data-driven-personal-finance-d97671225f81?source=rss-7d14ec5b901a------2</link>
            <guid isPermaLink="false">https://medium.com/p/d97671225f81</guid>
            <category><![CDATA[personal-finance]]></category>
            <dc:creator><![CDATA[Armin Norouzi, Ph.D]]></dc:creator>
            <pubDate>Thu, 06 Aug 2026 02:06:44 GMT</pubDate>
            <atom:updated>2026-08-06T02:06:44.686Z</atom:updated>
            <content:encoded><![CDATA[<p>Hey friend,</p><p>I want to tell you about something I have been quietly building on nights and weekends, and then ask you one small favor at the end.</p><p>By day I am an AI engineer. I spend my hours debugging systems, reading logs, and hunting down the one variable that quietly breaks everything downstream. A few years ago it hit me that I was treating my own money nothing like that. No dashboard, no logs, just vibes and a vague hope it would work out. So I started treating my finances the way I treat a system. Measure it, model it, and stop guessing.</p><p>That turned into a website:</p><blockquote><a href="http://datadrivenpersonalfinance.com/">datadrivenpersonalfinance.com</a></blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*MRurqlUZxUHHNt9crqtkQQ.png" /></figure><p>It is a growing set of free calculators and simulators, plus a book I am writing called <em>Debugging Your Personal Finance</em>. Everything on the site runs entirely in your browser. Nothing you type is sent to a server or saved anywhere I can see.</p><blockquote>No signup, no dark patterns, no “enter your email to see your results.”</blockquote><p>The one I am proudest of is the Portfolio Analyzer:</p><blockquote><a href="https://datadrivenpersonalfinance.com/portfolio-analyzer/">https://datadrivenpersonalfinance.com/portfolio-analyzer/</a></blockquote><p>Most retirement calculators hand you a single confident number: “you will have $2.3 million.” That number is a guess wearing a suit, because real markets are random. So instead of one tidy lie, this one takes your actual asset mix and runs it through 2,000 simulated market paths, then shows you the whole range of what could happen. The good years, the ugly ones, and the median in between. You can even build your own mix and watch the uncertainty widen or shrink as you change it. It is the closest thing I know to looking at your financial future honestly, with the error bars left on.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*yNbhQlYWf-2gPvyYRcOJRA.png" /></figure><p>There is more on the site now. When you can actually retire. Rent versus buy. What a 1% fee quietly costs you over a lifetime. Your real hourly wage after the commute and the lattes. All free, all in your browser, no strings.</p><h3>Here is the small favor…</h3><p>I just started a free newsletter. Once a week I send one short, data-driven breakdown, the kind of thing I wish someone had shown me at 25. No fluff, no “10 money hacks that will shock you,” just one clear idea and the math underneath it. If that sounds like your thing, I would be honored if you subscribed:</p><blockquote><a href="https://datadrivenpersonalfinance.com/subscribe">https://datadrivenpersonalfinance.com/subscribe</a></blockquote><p>I also just launched a LinkedIn page for this work. If you would rather follow along there, come say hi and give it a follow:</p><blockquote><a href="http://linkedin.com/company/data-driven-personal-finance/"><em>linkedin.com/company/data-driven-personal-finance/</em></a></blockquote><blockquote><strong>And one honest note, because some of you found me here for the AI and engineering writing, not the money stuff. </strong>I get it, and I am not going anywhere on Medium. I will keep publishing the tech pieces right here, same as always.</blockquote><p>So if personal finance is not your lane, no hard feelings at all. Just keep following me on Medium and I will keep the engineering posts coming.</p><p>Either way, thank you for reading me over the years. It genuinely means a lot. Go poke at the simulator, try to break it, and tell me what you would want to see next.</p><p>Talk soon,</p><p>Armin</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d97671225f81" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Model Collapse: What Happens When AI Trains on AI-Generated Data]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://levelup.gitconnected.com/model-collapse-what-happens-when-ai-trains-on-ai-generated-data-42a91ac7f2f3?source=rss-7d14ec5b901a------2"><img src="https://cdn-images-1.medium.com/max/990/1*1Obi5C9Vt6INl5IeA8Dn3w.png" width="990"></a></p><p class="medium-feed-snippet">When a model is trained on its own outputs and the cycle repeats across generations, variance collapses. Standard deviation drops from&#x2026;</p><p class="medium-feed-link"><a href="https://levelup.gitconnected.com/model-collapse-what-happens-when-ai-trains-on-ai-generated-data-42a91ac7f2f3?source=rss-7d14ec5b901a------2">Continue reading on Level Up Coding »</a></p></div>]]></description>
            <link>https://levelup.gitconnected.com/model-collapse-what-happens-when-ai-trains-on-ai-generated-data-42a91ac7f2f3?source=rss-7d14ec5b901a------2</link>
            <guid isPermaLink="false">https://medium.com/p/42a91ac7f2f3</guid>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[llm]]></category>
            <category><![CDATA[ai-generated-content]]></category>
            <category><![CDATA[training-data]]></category>
            <dc:creator><![CDATA[Armin Norouzi, Ph.D]]></dc:creator>
            <pubDate>Mon, 03 Aug 2026 15:32:57 GMT</pubDate>
            <atom:updated>2026-08-03T15:32:57.815Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Mixture of Experts: How Sparse Gating Enables 1T-Parameter LLMs]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://pub.towardsai.net/mixture-of-experts-how-sparse-gating-enables-1t-parameter-llms-655d20b68eb8?source=rss-7d14ec5b901a------2"><img src="https://cdn-images-1.medium.com/max/1255/1*6j7XRi0-z_zFZ3IN_41Amg.png" width="1255"></a></p><p class="medium-feed-snippet">A 1-trillion-parameter model sounds impossible to run &#x2014; until you realize that most of those parameters are inactive for any given token&#x2026;</p><p class="medium-feed-link"><a href="https://pub.towardsai.net/mixture-of-experts-how-sparse-gating-enables-1t-parameter-llms-655d20b68eb8?source=rss-7d14ec5b901a------2">Continue reading on Towards AI »</a></p></div>]]></description>
            <link>https://pub.towardsai.net/mixture-of-experts-how-sparse-gating-enables-1t-parameter-llms-655d20b68eb8?source=rss-7d14ec5b901a------2</link>
            <guid isPermaLink="false">https://medium.com/p/655d20b68eb8</guid>
            <dc:creator><![CDATA[Armin Norouzi, Ph.D]]></dc:creator>
            <pubDate>Mon, 03 Aug 2026 14:01:03 GMT</pubDate>
            <atom:updated>2026-08-03T14:01:03.492Z</atom:updated>
        </item>
    </channel>
</rss>