<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Requisite Variety on Eigenform Continuous Learning Tool Tips</title><link>https://www.eigenform.ai/continuous-learning-tool-tips/tags/requisite-variety/</link><description>Recent content in Requisite Variety on Eigenform Continuous Learning Tool Tips</description><generator>Hugo</generator><language>en-US</language><lastBuildDate>Tue, 17 Nov 2026 09:00:00 +0800</lastBuildDate><atom:link href="https://www.eigenform.ai/continuous-learning-tool-tips/tags/requisite-variety/index.xml" rel="self" type="application/rss+xml"/><item><title>RLHF Alternatives: Requisite Variety in Agent Training</title><link>https://www.eigenform.ai/continuous-learning-tool-tips/rlhf-alternatives/</link><pubDate>Tue, 17 Nov 2026 09:00:00 +0800</pubDate><guid>https://www.eigenform.ai/continuous-learning-tool-tips/rlhf-alternatives/</guid><description>&lt;figure class="article-image"&gt;
 &lt;picture&gt;
 &lt;source type="image/webp" srcset="https://www.eigenform.ai/continuous-learning-tool-tips/rlhf-alternatives/cover-rlhf-alternatives_hu_118830bdebea77a0.webp 480w, https://www.eigenform.ai/continuous-learning-tool-tips/rlhf-alternatives/cover-rlhf-alternatives_hu_3c08c8dd5691d288.webp 800w, https://www.eigenform.ai/continuous-learning-tool-tips/rlhf-alternatives/cover-rlhf-alternatives_hu_49d2a38fe5ee60e4.webp 1200w" sizes="(min-width: 900px) 800px, 100vw"&gt;
 &lt;source type="image/webp" srcset="https://www.eigenform.ai/continuous-learning-tool-tips/rlhf-alternatives/cover-rlhf-alternatives_hu_8827b3b6e078b8af.webp 480w, https://www.eigenform.ai/continuous-learning-tool-tips/rlhf-alternatives/cover-rlhf-alternatives_hu_3f0f292e5bcf71d4.webp 800w, https://www.eigenform.ai/continuous-learning-tool-tips/rlhf-alternatives/cover-rlhf-alternatives_hu_a18e85e47997c666.webp 1200w" sizes="(min-width: 900px) 800px, 100vw"&gt;
 &lt;img src="https://www.eigenform.ai/continuous-learning-tool-tips/rlhf-alternatives/cover-rlhf-alternatives_hu_3f0f292e5bcf71d4.webp" alt="Requisite variety in agent training as an etched parchment engraving: on the left, a narrow target gauge held by a large machine of preference scrolls, a reward model and a tether to a reference weight, with an inset curve that peaks and falls; on the right, a press, test tubes, a file drawer and scales whose results pass through a sieve into a tray and then into a small LoRA adapter on a model cube" title="A low-dimensional target needs a large reward apparatus; a rich environment lets a simple update rule do the job" width="1672" height="941" loading="eager" fetchpriority="high" decoding="async" class="zoomable" data-full-url="https://www.eigenform.ai/continuous-learning-tool-tips/rlhf-alternatives/cover-rlhf-alternatives.webp"&gt;
 &lt;/picture&gt;&lt;figcaption&gt;A low-dimensional target needs a large reward apparatus; a rich environment lets a simple update rule do the job&lt;/figcaption&gt;&lt;/figure&gt;&lt;p&gt;&lt;strong&gt;TL;DR:&lt;/strong&gt; Ashby&amp;rsquo;s law of requisite variety states that only variety can destroy variety: a regulator holding an outcome in a narrow band needs at least as many distinct responses as the disturbances it faces. RLHF (Reinforcement Learning from Human Feedback) holds a high-capacity LLM (large language model) to a low-dimensional target (helpful, harmless, honest) by building a high-variety supervisory apparatus around it, and most of its known failure modes are that apparatus running out of variety. The RLHF alternatives differ mainly in where they relocate the variety, into preference data with DPO (Direct Preference Optimisation), a judge model with RLAIF (Reinforcement Learning from AI Feedback), a verifier with RLVR (Reinforcement Learning with Verifiable Rewards) or the environment itself (filtered self-training), and the last choice is what lets the update rule shrink to SFT (supervised fine-tuning) on whatever survived.&lt;/p&gt;</description></item></channel></rss>