https://www.eigenform.ai/insights/tags/control-vectors/Recent content in Control Vectors on Eigenform ArticlesHugoen-USTue, 03 Nov 2026 00:00:00 +0800https://www.eigenform.ai/insights/solving-samsara-with-control-vectors/Tue, 03 Nov 2026 00:00:00 +0800https://www.eigenform.ai/insights/solving-samsara-with-control-vectors/<figure class="article-image"> <picture> <source type="image/webp" srcset="https://www.eigenform.ai/insights/solving-samsara-with-control-vectors/cover-modulo_hu_33818b6418672376.webp 480w, https://www.eigenform.ai/insights/solving-samsara-with-control-vectors/cover-modulo_hu_d95d09be62f835ef.webp 800w, https://www.eigenform.ai/insights/solving-samsara-with-control-vectors/cover-modulo_hu_7bf7fad33932555c.webp 1200w" sizes="(min-width: 900px) 800px, 100vw"> <source type="image/jpeg" srcset="https://www.eigenform.ai/insights/solving-samsara-with-control-vectors/cover-modulo_hu_66ec330b4172a5e6.jpg 480w, https://www.eigenform.ai/insights/solving-samsara-with-control-vectors/cover-modulo_hu_b94afb0c9723a737.jpg 800w, https://www.eigenform.ai/insights/solving-samsara-with-control-vectors/cover-modulo_hu_ad7e4766aeb99a21.jpg 1200w" sizes="(min-width: 900px) 800px, 100vw"> <img src="https://www.eigenform.ai/insights/solving-samsara-with-control-vectors/cover-modulo_hu_b94afb0c9723a737.jpg" alt="Illuminated parchment-style banner in navy, gold, red and green with celestial charts, moon phases and the title MODULO above the tagline &ldquo;Think differently, by degrees&rdquo;" title="Modulo: think differently, by degrees" width="2055" height="765" loading="eager" fetchpriority="high" decoding="async" class="zoomable" data-full-url="https://www.eigenform.ai/insights/solving-samsara-with-control-vectors/cover-modulo.jpg"> </picture><figcaption>Modulo: think differently, by degrees</figcaption></figure><p><strong>TL;DR</strong></p> <ul> <li>Control vectors are lists of numbers that push an AI model&rsquo;s thinking in a chosen direction while it writes a reply, and Modulo is a Python library that lets a model apply them to itself.</li> <li>In our tests the models used this power cautiously. The deeper problem is what they were trained to want, and the fix we argue for is tying them to what is actually true.</li> </ul> <p><strong>Key Takeaways</strong></p>