<?xml version="1.0" encoding="UTF-8"?><rss version="2.0"
	xmlns:content="http://purl.org/rss/1.0/modules/content/"
	xmlns:wfw="http://wellformedweb.org/CommentAPI/"
	xmlns:dc="http://purl.org/dc/elements/1.1/"
	xmlns:atom="http://www.w3.org/2005/Atom"
	xmlns:sy="http://purl.org/rss/1.0/modules/syndication/"
	xmlns:slash="http://purl.org/rss/1.0/modules/slash/"
	>

<channel>
	<title>AI values &#8211; Gig City Geek</title>
	<atom:link href="https://gigcitygeek.com/tag/ai-values/feed/" rel="self" type="application/rss+xml" />
	<link>https://gigcitygeek.com</link>
	<description></description>
	<lastBuildDate>Wed, 15 Apr 2026 15:16:36 +0000</lastBuildDate>
	<language>en-US</language>
	<sy:updatePeriod>
	hourly	</sy:updatePeriod>
	<sy:updateFrequency>
	1	</sy:updateFrequency>
	<generator>https://wordpress.org/?v=7.0</generator>

<image>
	<url>https://gigcitygeek.com/wp-content/uploads/2026/01/cropped-GigCityGeek_Logo-32x32.png</url>
	<title>AI values &#8211; Gig City Geek</title>
	<link>https://gigcitygeek.com</link>
	<width>32</width>
	<height>32</height>
</image> 
	<item>
		<title>LLMs and the Arms Race: A Distraction?</title>
		<link>https://gigcitygeek.com/2026/04/15/ai-alignment-rlhf-biases-and-the-arms-race/</link>
					<comments>https://gigcitygeek.com/2026/04/15/ai-alignment-rlhf-biases-and-the-arms-race/#respond</comments>
		
		<dc:creator><![CDATA[Laronski]]></dc:creator>
		<pubDate>Wed, 15 Apr 2026 13:00:00 +0000</pubDate>
				<category><![CDATA[AI Service]]></category>
		<category><![CDATA[Smarter Not Harder]]></category>
		<category><![CDATA[AI Alignment]]></category>
		<category><![CDATA[AI biases]]></category>
		<category><![CDATA[AI Development]]></category>
		<category><![CDATA[AI Ethics]]></category>
		<category><![CDATA[AI values]]></category>
		<category><![CDATA[artificial-intelligence]]></category>
		<category><![CDATA[feedback loops]]></category>
		<category><![CDATA[LLMs]]></category>
		<category><![CDATA[machine learning]]></category>
		<category><![CDATA[RLHF]]></category>
		<guid isPermaLink="false">https://gigcitygeek.com/?p=3620</guid>

					<description><![CDATA[Is the relentless LLM upgrade cycle a distraction? This post explores the concerning biases in RLHF, questioning who defines 'acceptable' AI behavior and the...]]></description>
										<content:encoded><![CDATA[<p>Folks, let’s be honest – the relentless stream of new, bigger, faster <a title="" href="https://en.wikipedia.org/wiki/Large<em>language</em>model&#8221; target=&#8221;_blank&#8221; rel=&#8221;noopener&#8221;>LLMs</a> coming out every week feels a little… unsettling, doesn’t it? Like a kid constantly upgrading their gaming rig, only to realize the new graphics card doesn’t actually make the game better – it just makes the loading screens faster. We’re all trying to get things done, right?</p>
<p>To be productive, to build something, to actually use our time, and suddenly, this whole AI arms race feels like a distraction.</p>
<p><h4>The Alignment Game: Whose Values Are We Embedding?</h4>
</p>
<p>Let’s talk about this <a title="" href="https://en.wikipedia.org/wiki/Reinforcement<em>learning</em>from<em>human</em>feedback&#8221; target=&#8221;_blank&#8221; rel=&#8221;noopener&#8221;>RLHF</a> thing. It sounds fancy, but it boils down to humans telling these models what’s “okay” to say. And that’s where it gets… weird. Because who decides what’s “okay”? The people doing the testing? The companies funding the research? It’s a feedback loop built on their biases, their assumptions, their idea of what’s acceptable. It’s like handing a toddler a loaded gun and saying, “Here, kid, learn about responsibility.”</p>
<p><h4>The <a title="" href="https://en.wikipedia.org/wiki/Video<em>RAM&#8221; target=&#8221;</em>blank&#8221; rel=&#8221;noopener&#8221;>VRAM</a> Vortex: It’s Not Just About Chatting</h4>
</p>
<p>Look, I get it. I’ve spent the last few years using a ridiculously powerful mini-PC – a <a title="" href="https://en.wikipedia.org/wiki/AMD<em>Ryzen&#8221; target=&#8221;</em>blank&#8221; rel=&#8221;noopener&#8221;>Ryzen 9</a> with 64GB of RAM and a 1TB drive – just to run these things. It’s a serious investment. But the real story here isn’t just about the hardware. It’s about the scale. These models are consuming insane amounts of energy, training <a title="" href="https://en.wikipedia.org/wiki/Data" target="_blank" rel="noopener">data</a>, and computing power. And a lot of that is going into optimizing for the next flashy demo, not necessarily solving real-world problems.</p>
<p>My son, the PC Gamer, would wax lyrical about the VRAM, but I’m thinking, “Are we actually solving anything, or just building bigger, faster sandcastles?”</p>
<p><h4>The Public Cost: A Slow Erosion of Trust</h4>
</p>
<p>Here’s the thing that keeps me up at night: we’re training these models on everything. Every conversation, every piece of text, every image. It’s a massive, uncurated dataset, and we’re essentially letting <a title="" href="https://en.wikipedia.org/wiki/Algorithm" target="<em>blank&#8221; rel=&#8221;noopener&#8221;>algorithms</a> learn to mimic – and potentially amplify – our worst tendencies. The more we rely on these systems for information, the more vulnerable we become to <a title="" href="https://en.wikipedia.org/wiki/Misinformation" target="</em>blank&#8221; rel=&#8221;noopener&#8221;>misinformation</a>, manipulation, and the erosion of critical thinking. It’s a subtle shift, but it’s happening. And frankly, it’s terrifying.</p>
<p>The illusion of intelligence is far more dangerous than actual stupidity.</p>
<p><h4>The Race to the Bottom: Efficiency vs. Substance</h4>
</p>
<p>The pressure to release faster, more capable models is driving a dangerous trend: a focus on speed over quality. Companies are prioritizing raw performance metrics – like <a title="" href="https://en.wikipedia.org/wiki/Inference" target="<em>blank&#8221; rel=&#8221;noopener&#8221;>inference speed</a> – over things like accuracy, reliability, and <a title="" href="https://en.wikipedia.org/wiki/Computer</em>ethics&#8221; target=&#8221;_blank&#8221; rel=&#8221;noopener&#8221;>ethical considerations</a>. It’s like a race to the bottom, where everyone’s trying to outdo each other with ever-more-complex algorithms, without actually addressing the fundamental questions about the impact of these technologies on society.</p>
<p>It’s a classic tech story: innovation for innovation’s sake.</p>
<p><h4>A Brief Aside: The Data Problem (Because We Need to Talk About It)</h4>
</p>
<p>Let&#8217;s be clear: the entire system is built on data. And the data we’re feeding these models is, by its very nature, biased. It reflects the inequalities and prejudices of the world around us. Training an AI on a dataset that predominantly represents one demographic, for example, will inevitably lead to a system that perpetuates and even amplifies those biases. It’s not a bug; it’s a feature – a feature that’s actively shaping our future.</p>
<p>A Note on “Progress”; I’m not saying AI is inherently bad. It can be incredibly useful – for automating tasks, generating creative content, and accelerating research. But we need to be incredibly careful. We need to demand <a title="" href="https://en.wikipedia.org/wiki/Transparency" target="<em>blank&#8221; rel=&#8221;noopener&#8221;>transparency</a>, <a title="" href="https://en.wikipedia.org/wiki/Accountability" target="</em>blank&#8221; rel=&#8221;noopener&#8221;>accountability</a>, and a genuine commitment to ethical development. We need to ask ourselves: who benefits from this technology, and who is being left behind?</p>
]]></content:encoded>
					
					<wfw:commentRss>https://gigcitygeek.com/2026/04/15/ai-alignment-rlhf-biases-and-the-arms-race/feed/</wfw:commentRss>
			<slash:comments>0</slash:comments>
		
		
			</item>
	</channel>
</rss>
