<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Uk on Step Into Dev</title><link>https://stepinto.dev/tags/uk/</link><description>Recent content in Uk on Step Into Dev</description><generator>Hugo</generator><language>en-gb</language><lastBuildDate>Wed, 17 Jun 2026 00:00:00 +0000</lastBuildDate><atom:link href="https://stepinto.dev/tags/uk/index.xml" rel="self" type="application/rss+xml"/><item><title>Compute Isn't the Constraint</title><link>https://stepinto.dev/posts/compute-isnt-the-constraint/</link><pubDate>Wed, 17 Jun 2026 00:00:00 +0000</pubDate><guid>https://stepinto.dev/posts/compute-isnt-the-constraint/</guid><description>&lt;p&gt;After &lt;a href="https://stepinto.dev/posts/when-the-api-gets-switched-off/"&gt;When the API Gets Switched Off&lt;/a&gt; went up, Bram De Buyser, an old friend, wrote back and put his finger on a line of thought I&amp;rsquo;d neglected. The frontier-lab ambition got one paragraph and a vague nod to &amp;ldquo;a serious compute commitment&amp;rdquo;, and that nod is where the strategy is thinnest. His point, roughly:&lt;/p&gt;
&lt;blockquote&gt;
&lt;p&gt;The Sovereign AI fund is primarily about compute access, but labs like Anthropic, OpenAI and Mistral aren&amp;rsquo;t pumping out better models just because of compute access. They&amp;rsquo;re putting in core research the others don&amp;rsquo;t have yet, and that requires research staff and experimenting, not just training or inference compute. If the UK wants to compete on the frontier, it needs to come up with the actual frontier, not just train models based on papers from 2022 that the big labs already consider obsolete.&lt;/p&gt;</description></item><item><title>When the API Gets Switched Off</title><link>https://stepinto.dev/posts/when-the-api-gets-switched-off/</link><pubDate>Sun, 14 Jun 2026 00:00:00 +0000</pubDate><guid>https://stepinto.dev/posts/when-the-api-gets-switched-off/</guid><description>&lt;p&gt;On 13 June, Anthropic disabled Claude Fable 5 and Mythos 5 for every customer outside the United States. Not deprecated, not rate-limited, but disabled. US national security authorities raised concerns days after the public release, and the company wrote that &amp;ldquo;the net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance.&amp;rdquo;&lt;/p&gt;
&lt;p&gt;Fable 5 was only generally available for about three days before it got pulled, so it&amp;rsquo;s not as though it was load-bearing in anyone&amp;rsquo;s production stack. It was available in mine, in CrossMoore EngineCore, but sitting there as a capability while we learned how it behaved, not running anything that mattered. The shape of it is what counts though. No migration window, no deprecation notice. One evening the calls just came back as an error.&lt;/p&gt;</description></item></channel></rss>