<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Lucas Boscaini.com</title><link>https://lucasboscaini.com/</link><description>Recent content on Lucas Boscaini.com</description><generator>Hugo</generator><language>en-us</language><lastBuildDate>Fri, 25 Sep 2026 00:00:00 +0000</lastBuildDate><atom:link href="https://lucasboscaini.com/index.xml" rel="self" type="application/rss+xml"/><item><title>Post-Training Free Task Vectors</title><link>https://lucasboscaini.com/posts/post-training-free-task-vectors/</link><pubDate>Fri, 25 Sep 2026 00:00:00 +0000</pubDate><guid>https://lucasboscaini.com/posts/post-training-free-task-vectors/</guid><description>&lt;h1 id="training-free-task-vectors-for-controlling-llm-behavior"&gt;Training-Free Task Vectors for Controlling LLM Behavior&lt;/h1&gt;&#10;&lt;p&gt;I am excited to share our latest research, &lt;strong&gt;&amp;ldquo;Training-Free Task Vectors for LLM Behavioral Control,&amp;rdquo;&lt;/strong&gt; which has been accepted at &lt;strong&gt;NeurIPS 2026&lt;/strong&gt;!&lt;/p&gt;&#10;&lt;p&gt;In this paper, co-authored with Gabriel J. Perin, Lucas Boscaini, André Araujo, and Nina S. T. Hirata, we tackle a fundamental challenge in post-training model editing: &lt;em&gt;How can we permanently steer LLM behaviors without the heavy computational cost of fine-tuning?&lt;/em&gt;&lt;/p&gt;&#10;&lt;hr&gt;&#10;&lt;h2 id="the-problem-the-high-cost-of-classical-task-vectors"&gt;The Problem: The High Cost of Classical Task Vectors&lt;/h2&gt;&#10;&lt;p&gt;Task vectors have emerged as a powerful paradigm for model editing. Traditionally, a task vector is constructed by taking the parameter-level difference between a fine-tuned model and its base checkpoint ( $\Delta\theta = \theta_{\text{fine-tuned}} - \theta_{\text{base}}$ ).&lt;/p&gt;</description></item><item><title>AI Security Development</title><link>https://lucasboscaini.com/talks/ai-security-development-painel/</link><pubDate>Tue, 08 Sep 2026 00:00:00 +0000</pubDate><guid>https://lucasboscaini.com/talks/ai-security-development-painel/</guid><description>&lt;p&gt;Broad discussion on AI, security, and software development, with invited speakers from industry and academia.&lt;/p&gt;&#10;&lt;p&gt;Presented at CBSoft: &lt;a href="https://cbsoft.sbc.org.br/2026/pt/workshops/event/#panel-discussion-aisecdev"&gt;https://cbsoft.sbc.org.br/2026/pt/workshops/event/#panel-discussion-aisecdev&lt;/a&gt;&lt;/p&gt;</description></item><item><title>About</title><link>https://lucasboscaini.com/about/</link><pubDate>Mon, 02 Feb 2026 00:00:00 +0000</pubDate><guid>https://lucasboscaini.com/about/</guid><description>&lt;h2 id="who-am-i-"&gt;Who am I ?&lt;/h2&gt;&#10;&lt;p&gt;&lt;strong&gt;I am Lucas Boscaini, a technologist that has been leading the research, discovery, and creation of new technologies, with experience in big techs and startups.&lt;/strong&gt;&lt;/p&gt;&#10;&lt;p&gt;I am a builder and strategist focused on the intersection of technical innovation and practical execution. Throughout my career, I have navigated the high-scale environments of Big Tech and the high-velocity world of startups. This dual perspective has shaped my belief that great technology isn&amp;rsquo;t just about elegant code—it’s about solving the right problems at the right time.&lt;/p&gt;</description></item><item><title>Ransomware prevention and Good practices</title><link>https://lucasboscaini.com/talks/ransomware-prevention-and-good-practices/</link><pubDate>Wed, 02 Apr 2025 00:00:00 +0000</pubDate><guid>https://lucasboscaini.com/talks/ransomware-prevention-and-good-practices/</guid><description>&lt;p&gt;Talk in PT-BR - Transmited via Youtube: &lt;a href="https://www.youtube.com/watch?v=3Tr3fM6U_1k"&gt;https://www.youtube.com/watch?v=3Tr3fM6U_1k&lt;/a&gt;&lt;/p&gt;</description></item><item><title>AI on the Rise - Security Challenges in the Era of Artificial Intelligence</title><link>https://lucasboscaini.com/talks/ai-on-the-rise---security-challenges-in-the-era-of-artificial-intelligence/</link><pubDate>Tue, 17 Sep 2024 00:00:00 +0000</pubDate><guid>https://lucasboscaini.com/talks/ai-on-the-rise---security-challenges-in-the-era-of-artificial-intelligence/</guid><description>&lt;p&gt;Artificial Intelligence is transforming the world at an accelerated pace, opening doors to incredible innovations but also bringing to light new and complex security challenges. In this talk, we will explore how the rapid adoption of AI is impacting the global landscape, from the creation of sophisticated cyberattacks to infiltrations in critical systems. We will discuss the crucial role of security experts in developing new technologies and strategies to ensure that AI integration is done responsibly and sustainably, protecting individuals, organizations, and society as a whole.&lt;/p&gt;</description></item></channel></rss>