Matej Ľach ✅
MatejLach@social.matej-lach.me
<p>Free software enthusiast, <a href="https://social.matej-lach.me/tags/golang" class="mention hashtag" rel="tag">#<span>golang</span></a>, <a href="https://social.matej-lach.me/tags/rustlang" class="mention hashtag" rel="tag">#<span>rustlang</span></a>, <a href="https://social.matej-lach.me/tags/swiftlang" class="mention hashtag" rel="tag">#<span>swiftlang</span></a> . Working on a question/answer <a href="https://social.matej-lach.me/tags/ActivityPub" class="mention hashtag" rel="tag">#<span>ActivityPub</span></a> server. <a href="https://social.matej-lach.me/tags/systemd" class="mention hashtag" rel="tag">#<span>systemd</span></a> aficionado :-)</p>
Posts
-
View post
The U.S. is intent on limiting GPU exports to #China precisely to limit releases of open models that compete with proprietary American ones, like the DeepSeek R1. The U.S. only believes in the free market until there&#39;s competition.
-
View post
If this doesn&#39;t say &#39;we&#39;re greedy and can&#39;t compete&#39;, I don&#39;t know what does.
-
View post
#OpenAI complaining about someone maybe using its &#39;outputs&#39; while training on copyrighted data is fucking rich.
-
View post
Translation: American officials strongly believe in free market competition for Venezuela, Cuba or Nicaragua, not for the U.S. however.
-
View post
I really appreciate high profile podcasters putting out quality content week after week. /s
-
View post
Wait, so #DeepSeek releases an open model and the response of U.S. companies is to not only keep their models closed but to debate whether they should even publish research papers now? The implication being that no one in #China could possibly have figured this stuff out without the allmighty U.S. publishing research papers? You know what would&#39;ve been a badass response? Opening your models and collaborating on an even better one, maybe even with DeepSeek. Instead we get this orienta...
-
View post
#France is apparently pledging to build its own Stargate and invest massively in data centers and power for AI training. Instead of learning from #DeepSeek that it may not all be about brute force compute but rather that the next big leap is in going back to the basics and OPTIMIZING, OPTIMIZING, OPTIMIZING - no, instead we&#39;re going to waste huge amounts on building data centers to train AI that doesn&#39;t do anything to justify its costs just so that we don&#39;t hurt our ego...
-
View post
What&#39;s up with laptops being stuck at 1TB storage for the past several years?
-
View post
It&#39;s good to see that even in some notorious podcast circles, smugness and ignorance is slowly being recognised for what it is and that it&#39;s no longer enough to have a British accent to pass for an intellectual.
-
View post
Ok, this is just bonkers. #xAI hasn&#39;t launched a single &#39;original&#39; or &#39;breakthrough product and there&#39;s no reason to believe it ever will. It has a copycat model that, despite its advantage of having unlimited access to Twitter data to train on, is at best &#39;average&#39; and at worst gets beaten by open, locally runnable models like QwQ/R1 It&#39;s not &#39;best in class&#39; in anything and models like Gemini 2.5 Pro just set it...
-
View post
The idea that a country like #China could ever be seriously hindered by not being able to buy the latest from #NVidia shows how the West still thinks of itself as uniquely capable of &#39;hard engineering&#39; really makes one wonder if economic decisions (or any decisions for that matter) are ever based on hard facts, rich data and smart projections, or is it more often that we&#39;d like to imagine just someone&#39;s feelings?
-
View post
MAKE AMERICAN EXPORTS GREAT AGAIN (one extortion at a time)