<?xml version="1.0" encoding="utf-8" standalone="yes"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:content="http://purl.org/rss/1.0/modules/content/">
  <channel>
    <title>Gpt on Taras&#39; Blog on AI, Perf, Hacks</title>
    <link>/tags/gpt/</link>
    <description>Recent content in Gpt on Taras&#39; Blog on AI, Perf, Hacks</description>
    <image>
      <title>Taras&#39; Blog on AI, Perf, Hacks</title>
      <url>/images/papermod-cover.png</url>
      <link>/images/papermod-cover.png</link>
    </image>
    <generator>Hugo -- 0.147.0</generator>
    <language>en</language>
    <copyright>Taras Glek</copyright>
    <lastBuildDate>Mon, 07 Aug 2023 10:45:27 +0300</lastBuildDate>
    <atom:link href="/tags/gpt/index.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>Overlooked on HN: Discovering High-quality Technical Content</title>
      <link>/posts/overlooked-on-hn-discovering-technical-content/</link>
      <pubDate>Mon, 07 Aug 2023 10:45:27 +0300</pubDate>
      <guid>/posts/overlooked-on-hn-discovering-technical-content/</guid>
      <description>&lt;p&gt;I&amp;rsquo;m gonna start a column on cool blog posts I found, that got 0 or minimal traction. I suspect I will also have no traction doing that 🤦‍♂️.&lt;/p&gt;
&lt;h2 id=&#34;the-problem&#34;&gt;The Problem&lt;/h2&gt;
&lt;p&gt;I really enjoy thoughtful writing on deep technical problems. It&amp;rsquo;s even better when one sees thoughtful comments, that further contribute new directions to throughts presented. &lt;a href=&#34;https://news.ycombinator.com/&#34;&gt;HackerNews&lt;/a&gt; is where most of that writing tends to land. Unfortunately it tends to not do well vs trendy, click-baity, etc content. Twitter is even worse.&lt;/p&gt;
&lt;p&gt;First, a blog post on my tooling for reading HN.&lt;/p&gt;</description>
    </item>
    <item>
      <title>Always Bet on Geohot: Tinygrad Will Win</title>
      <link>/posts/always-bet-on-geohot/</link>
      <pubDate>Thu, 25 May 2023 11:23:14 +0300</pubDate>
      <guid>/posts/always-bet-on-geohot/</guid>
      <description>&lt;p&gt;George Holtz (geohot) just raised &lt;a href=&#34;https://geohot.github.io//blog/jekyll/update/2023/05/24/the-tiny-corp-raised-5M.html&#34;&gt;$5M&lt;/a&gt; for tinygrad and plans to sell a $15,000 Machine Learning AMD Epyc box with &lt;a href=&#34;https://www.amd.com/en/products/graphics/amd-radeon-rx-7900xt&#34;&gt;AMD 7900 XT&lt;/a&gt; video cards. At first glance, this seems like a really risky investment into a low-margin commodity hardware company. AMD can&amp;rsquo;t compete with NVidia, so how can we expect geohot to compete with NVidia on AMD hardware?&lt;/p&gt;
&lt;p&gt;Well, I&amp;rsquo;ve been following geohot for a long while. He delivers what he promises, but does it years later and on different hardware than initially promised :) Here&amp;rsquo;s the timeline of my evolution alongside George that informs my conviction:&lt;/p&gt;</description>
    </item>
    <item>
      <title>A Tale of PGVector and Friends: The Joy Of Software Naming</title>
      <link>/posts/a-tale-of-pgvector-and-friends/</link>
      <pubDate>Tue, 23 May 2023 18:20:05 +0300</pubDate>
      <guid>/posts/a-tale-of-pgvector-and-friends/</guid>
      <description>&lt;p&gt;Ah, the world of software! A place where creativity and collaboration come together to create amazing projects. But sometimes, this world can be a bit&amp;hellip; confusing. Let me tell you a story about the frustration of open source naming, featuring our protagonist: PGVector.&lt;/p&gt;
&lt;p&gt;Once upon a time, I though that pgvector was a &lt;a href=&#34;https://postgresql.org/&#34;&gt;postgresql&lt;/a&gt; extension. And that langchain did not support SQL storage out of the box. After reading the source, turned out there were three things named PGVector:&lt;/p&gt;</description>
    </item>
    <item>
      <title>GPT-ARIA: Integrating GPT-3 with Browser Using Accessibility API</title>
      <link>/posts/gpt-aria-experiment/</link>
      <pubDate>Thu, 30 Mar 2023 15:09:04 +0200</pubDate>
      <guid>/posts/gpt-aria-experiment/</guid>
      <description>&lt;p&gt;&lt;a href=&#34;https://github.com/thegpvc/gpt-aria&#34;&gt;https://github.com/thegpvc/gpt-aria&lt;/a&gt;&lt;/p&gt;
&lt;p&gt;This was the most interesting prototype I’ve explored in my career so far, and I want to thank &lt;a href=&#34;https://www.thegp.com/&#34;&gt;TheGP&lt;/a&gt; for sponsoring this project. Also thanks to &lt;a href=&#34;https://github.com/chugai&#34;&gt;Oleksandr Chugai&lt;/a&gt; who I partnered with on implementation and &lt;a href=&#34;https://twitter.com/bencmejla&#34;&gt;Ben Cmejla&lt;/a&gt; who helped with prompting.&lt;/p&gt;
&lt;h2 id=&#34;inspiration&#34;&gt;Inspiration&lt;/h2&gt;
&lt;p&gt;I’ve been thinking about browser automation since the early days of my career when I worked at Firefox, where I got to work on performance problems and built things like Firefox Telemetry (Hi, HN haters!). With the rise of large language models (LLMs)—and OpenAI’s recent launch of ChatGPT plugins — things are getting exciting around user agents again. My favorite demo so far is &lt;a href=&#34;https://github.com/nat/natbot&#34;&gt;natbot&lt;/a&gt;, Nat Friedman’s GPT-3-powered browser agent that he launched last October.&lt;/p&gt;</description>
    </item>
  </channel>
</rss>
