<?xml version="1.0" encoding="utf-8" standalone="yes"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
	<channel>
		<title>Home on things I do</title>
		<link>https://grokked.it/</link>
		<description>Recent content in Home on things I do</description>
		<generator>Hugo</generator>
		<language>en</language>
		
		
		
		
			<lastBuildDate>Thu, 16 Jul 2026 00:00:00 +0000</lastBuildDate>
		
			<atom:link href="https://grokked.it/index.xml" rel="self" type="application/rss+xml" />
			<item>
				<title>After The Bubble - Snake I</title>
				<link>https://grokked.it/posts/after_the_bubble/</link>
				<pubDate>Thu, 16 Jul 2026 00:00:00 +0000</pubDate>
				<guid>https://grokked.it/posts/after_the_bubble/</guid>
				<description>&lt;p&gt;I&amp;rsquo;ve expanded the &lt;a href=&#34;https://grokked.it/posts/ggufun/&#34;  target=&#34;_blank&#34; rel=&#34;noopener&#34; &gt;GGUFun&lt;/a&gt; project: I started by generating a deterministic transformer in a GGUF file that always produced the same exact sentence, then made it the parodistic website &amp;ldquo;After the bubble&amp;rdquo;, a catalog of transformers packaged in GGUF files for after the AI bubble bursts, for all the ollama servers around the world that will be left unused.&lt;/p&gt;&#xA;&lt;p&gt;While the transformers I hand crafted with the help of coding agents are preposterous, I did learn a thing or two while doing it, while it&amp;rsquo;s important to understand that no transformer here contains any value that has been &amp;ldquo;learned&amp;rdquo;, only manually set.&lt;/p&gt;</description>
			</item>
			<item>
				<title>GGUFun</title>
				<link>https://grokked.it/posts/ggufun/</link>
				<pubDate>Sat, 11 Jul 2026 00:00:00 +0000</pubDate>
				<guid>https://grokked.it/posts/ggufun/</guid>
				<description>&lt;p&gt;Lately I was wondering if it would be possible to build a transformer inside a gguf file that, when loaded by ollama, would always answer with the same text, whatever the input.&lt;/p&gt;&#xA;&lt;p&gt;I don&amp;rsquo;t mean something like an actual LLM with a strongly worded system prompt, I mean an actual deterministic output.&lt;/p&gt;&#xA;&lt;p&gt;Turns out: it can be done, and it&amp;rsquo;s not even too complex. You zero out the impact of the attention heads and MLPs on the residual stream, and hide everything inside the unembedding matrix.&lt;/p&gt;</description>
			</item>
			<item>
				<title>Mech Interp - Garden Path Sentences</title>
				<link>https://grokked.it/posts/garden_path_sentences/</link>
				<pubDate>Thu, 09 Jul 2026 00:00:00 +0000</pubDate>
				<guid>https://grokked.it/posts/garden_path_sentences/</guid>
				<description>&lt;p&gt;Garden Path Sentences (GPS from here on) are sentences like this one: &amp;ldquo;The old man the boat.&amp;rdquo;&lt;/p&gt;&#xA;&lt;p&gt;Reading the sentence word by word, we initially parse &amp;ldquo;The old man&amp;rdquo; as a noun phrase meaning an aged male. But when we reach the end, we realize that &amp;ldquo;man&amp;rdquo; isn&amp;rsquo;t a noun at all — it&amp;rsquo;s a verb, in the sense of &amp;ldquo;to crew&amp;rdquo; or &amp;ldquo;to operate.&amp;rdquo; Only then do we backtrack and reinterpret the sentence as &amp;ldquo;the old [people] operate the boat.&amp;rdquo; I don&amp;rsquo;t think I know any examples in Italian.&lt;/p&gt;</description>
			</item>
			<item>
				<title>Parallel logger</title>
				<link>https://grokked.it/posts/parallel_logger/</link>
				<pubDate>Thu, 20 Nov 2025 00:00:00 +0000</pubDate>
				<guid>https://grokked.it/posts/parallel_logger/</guid>
				<description>&lt;p&gt;Today I was running 4 jobs in parallel in python using ThreadPoolExecutor. The logs were totally incomprehensible, since they were all sending to stdout.&lt;/p&gt;&#xA;&lt;p&gt;I decided I wanted to create something more readable, and I (for the first time!) found out about &lt;a href=&#34;https://docs.python.org/3/howto/curses.html&#34;  target=&#34;_blank&#34; rel=&#34;noopener&#34; &gt;Curses&lt;/a&gt;, a python library that lets you interact with the terminal (to be honest: I&amp;rsquo;ve been using curses with python for years, but I&amp;rsquo;ve been cursing even more with javascript, so I don&amp;rsquo;t think that counts)&lt;/p&gt;</description>
			</item>
			<item>
				<title>Nextpress</title>
				<link>https://grokked.it/posts/yet_another_python_framework/</link>
				<pubDate>Thu, 09 Oct 2025 00:00:00 +0000</pubDate>
				<guid>https://grokked.it/posts/yet_another_python_framework/</guid>
				<description>&lt;p&gt;After trying to implement as much &lt;code&gt;ExpressJS&lt;/code&gt; as possible in &lt;code&gt;FastAPI&lt;/code&gt;, I realized that it could not be done. Therefore, I built my own ASGI compliant web framework that offers me a similar experience as with &lt;code&gt;ExpressJS&lt;/code&gt;.&lt;/p&gt;</description>
			</item>
			<item>
				<title>FastAPI Route Middleware</title>
				<link>https://grokked.it/posts/fastapi_route_middleware/</link>
				<pubDate>Tue, 23 Sep 2025 00:00:00 +0000</pubDate>
				<guid>https://grokked.it/posts/fastapi_route_middleware/</guid>
				<description>&lt;p&gt;This is a another python project: it&amp;rsquo;s a library allows to use middlewares in FastAPI in a manner that, to me, reminds me of how Express.js handles middlewares.&lt;/p&gt;</description>
			</item>
			<item>
				<title>Let&#39;s get fiscal</title>
				<link>https://grokked.it/posts/lets_get_fiscal/</link>
				<pubDate>Mon, 25 Aug 2025 00:00:00 +0000</pubDate>
				<guid>https://grokked.it/posts/lets_get_fiscal/</guid>
				<description>&lt;p&gt;This is just an introduction. I first noticed that at an hotel where I stayed ~1 year ago a specific server was exposed on the same wifi network that was available for customers. It was not the only one, but it was by far the most interesting for me.&lt;/p&gt;&#xA;&lt;p&gt;Even after telling them that, noting has changed, and after 1 year they still had everything exposed for all customers to see and use (such as their printer).&lt;/p&gt;</description>
			</item>
			<item>
				<title>Serial Gaming [WIP]</title>
				<link>https://grokked.it/posts/serial_gaming/</link>
				<pubDate>Thu, 31 Jul 2025 00:00:00 +0000</pubDate>
				<guid>https://grokked.it/posts/serial_gaming/</guid>
				<description>&lt;p&gt;I&amp;rsquo;ve had a couple of Raspberry PI pico at home for years, mostly unused, but recently I&amp;rsquo;ve wanted to try something with one of them.&lt;/p&gt;&#xA;&lt;p&gt;I&amp;rsquo;ve created a simple TicTacToe game in javascript, and used the &lt;a href=&#34;https://developer.mozilla.org/en-US/docs/Web/API/Web_Serial_API&#34;  target=&#34;_blank&#34; rel=&#34;noopener&#34; &gt;WebSerial interface&lt;/a&gt;, that only works in Chrome, to make a pico play the game using some micropython code that I wrote.&lt;/p&gt;&#xA;&lt;p&gt;I then wrote and tested different AIs, all running on the very limited hardware of a pico (264kB RAM and dual-core Arm Cortex-M0+ CPU): a random one, one that uses MinMax to decide the next move, one that learns each game which move is best given the current situation, and one based on a perceptron.&lt;/p&gt;</description>
			</item>
			<item>
				<title>Static TOTPs ebook generator</title>
				<link>https://grokked.it/posts/static_totps/</link>
				<pubDate>Wed, 09 Apr 2025 00:00:00 +0000</pubDate>
				<guid>https://grokked.it/posts/static_totps/</guid>
				<description>&lt;p&gt;We&amp;rsquo;ve been using TOTPs for 2FA for years, and I find them quite useful. I just find the rush of typing them in less than 30 seconds (or even less) to be too much pressure.&lt;/p&gt;&#xA;&lt;p&gt;I decided then to create a web app that simply scans a qrcode and uses it to generate an ebook (in epub format) containing a whole month of TOTPs, each with the exact time where that TOTP will be working.&lt;/p&gt;</description>
			</item>
			<item>
				<title>How to listen to Morning for free</title>
				<link>https://grokked.it/posts/ilpost/</link>
				<pubDate>Sat, 28 Sep 2024 00:00:00 +0000</pubDate>
				<guid>https://grokked.it/posts/ilpost/</guid>
				<description>&lt;p&gt;&amp;ldquo;Il Post&amp;rdquo; is an italian online newspaper to whom I have been subscribed for years. They also produce several podcast, some of which are for subscribers only, and can only be listened on the browser on desktop and on their app on mobile.&lt;/p&gt;</description>
			</item>
			<item>
				<title>Actual weight of olympic medals</title>
				<link>https://grokked.it/posts/olympic_medals/</link>
				<pubDate>Fri, 09 Aug 2024 00:00:00 +0000</pubDate>
				<guid>https://grokked.it/posts/olympic_medals/</guid>
				<description>&lt;p&gt;The Olympic Games in Paris bring back a question as old as time: should all medals be worth the same?&#xA;We all know that medals are big and shiny so that those who wear them can warn other people that they are way better than others in some disciplines.&lt;/p&gt;</description>
			</item>
			<item>
				<title>Magic Top Hat</title>
				<link>https://grokked.it/posts/magic_top_hat/</link>
				<pubDate>Thu, 11 Jul 2024 00:00:00 +0000</pubDate>
				<guid>https://grokked.it/posts/magic_top_hat/</guid>
				<description>&lt;p&gt;This is a small python project: it&amp;rsquo;s a library that uses an llm to generate a function at runtime based on the context in which the function was called.&lt;/p&gt;</description>
			</item>
			<item>
				<title>Manifestival</title>
				<link>https://grokked.it/posts/manifestival/manifestival/</link>
				<pubDate>Mon, 10 Jun 2024 00:00:00 +0000</pubDate>
				<guid>https://grokked.it/posts/manifestival/manifestival/</guid>
				<description>&lt;p&gt;&lt;a href=&#34;https://github.com/lucamattiazzi/manifestival&#34;  target=&#34;_blank&#34; rel=&#34;noopener&#34; &gt;Github&lt;/a&gt;&lt;/p&gt;&#xA;&lt;p&gt;This is a project that I did not finish in time, and will likely finish at a better time.&lt;/p&gt;&#xA;&lt;p&gt;In occasion of the 2024 european elections, I used two agents built using an LLM to impersonate the manifest of a party each. They were then asked a question (such as how to fix the economy) and they had to discuss.&lt;/p&gt;</description>
			</item>
			<item>
				<title>Kindertip</title>
				<link>https://grokked.it/posts/kindertip/</link>
				<pubDate>Sun, 14 Jan 2024 00:00:00 +0000</pubDate>
				<guid>https://grokked.it/posts/kindertip/</guid>
				<description>&lt;p&gt;I&amp;rsquo;ve started working on this project while visiting Sevilla in January 2024. At the time my daugher was 2 and still went to a kindergarten that started using an application to communicate live news with parents, Kindertap. On that app the teachers upload information on meals, news, contagious illnesses, and upload pictures.&lt;/p&gt;</description>
			</item>
			<item>
				<title>Costa Intro</title>
				<link>https://grokked.it/posts/costa_intro/</link>
				<pubDate>Thu, 09 Nov 2023 00:00:00 +0000</pubDate>
				<guid>https://grokked.it/posts/costa_intro/</guid>
				<description>&lt;p&gt;Quick and dirty nighttime project to analyze the intro length of every episode of &amp;ldquo;Morning&amp;rdquo; podcast from Francesco Costa, produced by Il Post.&lt;/p&gt;&#xA;&lt;p&gt;I downloaded every episode with beautifulsoup, then used whisper to quickly transcribe the episodes, even at the lowest quality.&lt;/p&gt;&#xA;&lt;p&gt;Used nltk to look for words similar to &amp;ldquo;morning cominciamo&amp;rdquo;, since it&amp;rsquo;s the keyword that the podcast author always used to end the introduction.&lt;/p&gt;&#xA;&lt;p&gt;Use jupyter + pandas + matplotlib to plot all the intro length through time.&lt;/p&gt;</description>
			</item>
	</channel>
</rss>
