<rss version="2.0" xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Hacker News: krapht</title><link>https://news.ycombinator.com/user?id=krapht</link><description>Hacker News RSS</description><docs>https://hnrss.org/</docs><generator>hnrss v2.1.1</generator><lastBuildDate>Thu, 06 Aug 2026 03:21:08 +0000</lastBuildDate><atom:link href="https://hnrss.org/user?id=krapht" rel="self" type="application/rss+xml"></atom:link><item><title><![CDATA[New comment by krapht in "Changes at Google DeepMind: Demis Hassabis from CEO to Chair, Jeff Dean departs"]]></title><description><![CDATA[
<p>social media is a completely different market though - since there are massive returns to scale, it's incredibly hard for a new entrant to break in.<p>model training and inference is the opposite. people switch LLMs like people change clothes in the morning. there are popular services (openrouter) that make moving as easy as changing a model string.</p>
]]></description><pubDate>Wed, 05 Aug 2026 16:41:31 +0000</pubDate><link>https://news.ycombinator.com/item?id=49185309</link><dc:creator>krapht</dc:creator><comments>https://news.ycombinator.com/item?id=49185309</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=49185309</guid></item><item><title><![CDATA[New comment by krapht in "Changes at Google DeepMind: Demis Hassabis from CEO to Chair, Jeff Dean departs"]]></title><description><![CDATA[
<p>Is Gemini really doomed? I'm still bullish on Google:
1) they have more free cash flow and capital than God due to the ads business
2) they have data - intent from web searches, youtube videos, google books and music
3) they have dedicated inference hardware<p>for all these reasons, is being 6 months behind the frontier actually a structural, long term disadvantage? some day the pace of improvement will slow, and google will vacuum up the market. they'll be able to compete with open-weight models just on pure cost advantage from their vertical integration</p>
]]></description><pubDate>Wed, 05 Aug 2026 16:38:27 +0000</pubDate><link>https://news.ycombinator.com/item?id=49185258</link><dc:creator>krapht</dc:creator><comments>https://news.ycombinator.com/item?id=49185258</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=49185258</guid></item><item><title><![CDATA[New comment by krapht in "DeepSeek-V4-Flash Update"]]></title><description><![CDATA[
<p>I think that's backwards. Anything that changes the performance of a model deserves a minor version bump. A new model has to be qualified before being pushed to production; but we don't get the choice here, just cross your fingers there are no regressions at all on all possible tasks the model might be asked to do.</p>
]]></description><pubDate>Fri, 31 Jul 2026 12:38:39 +0000</pubDate><link>https://news.ycombinator.com/item?id=49122354</link><dc:creator>krapht</dc:creator><comments>https://news.ycombinator.com/item?id=49122354</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=49122354</guid></item><item><title><![CDATA[New comment by krapht in "A $500 RL fine-tune of a 9B open model beat frontier models on catalog review"]]></title><description><![CDATA[
<p>News to me. What do you call mixture-of-experts and model distillation?</p>
]]></description><pubDate>Tue, 28 Jul 2026 16:22:34 +0000</pubDate><link>https://news.ycombinator.com/item?id=49086250</link><dc:creator>krapht</dc:creator><comments>https://news.ycombinator.com/item?id=49086250</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=49086250</guid></item><item><title><![CDATA[New comment by krapht in "Gemini last models: temperature, top_p, and top_k are deprecated and ignored"]]></title><description><![CDATA[
<p>ironically, any frontier LLM will easily generate a tutorial at any detail you like explaining what these are.<p>if don't have time for that, just know that these are technical parameters that affect how likely it is an llm will produce the same result after being asked the same question.</p>
]]></description><pubDate>Wed, 22 Jul 2026 01:58:47 +0000</pubDate><link>https://news.ycombinator.com/item?id=49000915</link><dc:creator>krapht</dc:creator><comments>https://news.ycombinator.com/item?id=49000915</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=49000915</guid></item><item><title><![CDATA[New comment by krapht in "Why I Left Google DeepMind"]]></title><description><![CDATA[
<p>It's more rare, and definitely less lucrative, but there are  opportunities in software engineering for non-profits, open-source, and smaller non-tech companies. If you've made your blood money and don't need top n% pay, it's something worth checking out.</p>
]]></description><pubDate>Wed, 15 Jul 2026 21:24:33 +0000</pubDate><link>https://news.ycombinator.com/item?id=48927200</link><dc:creator>krapht</dc:creator><comments>https://news.ycombinator.com/item?id=48927200</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=48927200</guid></item><item><title><![CDATA[New comment by krapht in "Designing and assembling my first PCB"]]></title><description><![CDATA[
<p>Slight tangent: a long time ago when I was into this hobby, Oshpark was usually the go-to instead of JLCPCB. What happened to them? Macrofab too if anyone's used them.</p>
]]></description><pubDate>Mon, 13 Jul 2026 03:33:02 +0000</pubDate><link>https://news.ycombinator.com/item?id=48887554</link><dc:creator>krapht</dc:creator><comments>https://news.ycombinator.com/item?id=48887554</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=48887554</guid></item><item><title><![CDATA[New comment by krapht in "An Introduction to YOLO26"]]></title><description><![CDATA[
<p>> The catch is that for image resolutions >=700x700pixels (most production usecases)<p>Citation needed? 2XL looks like you go up to 800x800 pixel inputs. This isn't the dealbreaker you say it is - all pipelines benefit from thoughtful crop and rescaling before going to inference.</p>
]]></description><pubDate>Tue, 23 Jun 2026 13:01:42 +0000</pubDate><link>https://news.ycombinator.com/item?id=48644337</link><dc:creator>krapht</dc:creator><comments>https://news.ycombinator.com/item?id=48644337</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=48644337</guid></item><item><title><![CDATA[New comment by krapht in "Has AI already killed self-help nonfiction books?"]]></title><description><![CDATA[
<p>More than just your commute. Any downtime where you are doing a mindless chore is prime podcast time.</p>
]]></description><pubDate>Tue, 16 Jun 2026 21:32:12 +0000</pubDate><link>https://news.ycombinator.com/item?id=48562437</link><dc:creator>krapht</dc:creator><comments>https://news.ycombinator.com/item?id=48562437</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=48562437</guid></item><item><title><![CDATA[New comment by krapht in "Fine-tuning an LLM to write docs like it's 1995"]]></title><description><![CDATA[
<p>I've heard of people using anythingllm for this purpose.<p>Basic rag is almost stupid in how easy it is, though. You grep for keywords, take the surrounding paragraph, then stuff it all into your llm prompt.<p>The next upgrade is to automate keyword extraction by putting your documents into a vector store and search by vector similarity.</p>
]]></description><pubDate>Fri, 05 Jun 2026 12:13:03 +0000</pubDate><link>https://news.ycombinator.com/item?id=48411348</link><dc:creator>krapht</dc:creator><comments>https://news.ycombinator.com/item?id=48411348</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=48411348</guid></item><item><title><![CDATA[New comment by krapht in "Expertise in the Age of AI"]]></title><description><![CDATA[
<p>Don't post generated comments or AI-edited comments. HN is for conversation between humans. <a href="https://news.ycombinator.com/newsguidelines.html#generated">https://news.ycombinator.com/newsguidelines.html#generated</a></p>
]]></description><pubDate>Fri, 29 May 2026 15:46:18 +0000</pubDate><link>https://news.ycombinator.com/item?id=48324710</link><dc:creator>krapht</dc:creator><comments>https://news.ycombinator.com/item?id=48324710</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=48324710</guid></item><item><title><![CDATA[New comment by krapht in "AI has a multiplying effect on existing technical skills"]]></title><description><![CDATA[
<p>I went and looked at the code. It's AI bots and a few confused trend followers all the way down. There's no way anything in there works.</p>
]]></description><pubDate>Fri, 22 May 2026 18:09:41 +0000</pubDate><link>https://news.ycombinator.com/item?id=48239312</link><dc:creator>krapht</dc:creator><comments>https://news.ycombinator.com/item?id=48239312</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=48239312</guid></item><item><title><![CDATA[New comment by krapht in "Hershey Bets on Agentic AI to Rethink $2B in Marketing Spend"]]></title><description><![CDATA[
<p><article excerpt><p>What AI actually does 
Mutinex has built what it describes as a “multi-agent system,” where each agent acts as a domain specialist. For example, one agent understands marketing econometrics, another understands competitive pricing theory, another diagnoses model failures.<p>By combining Tracer, which cleans and makes sense of Hershey’s data infrastructure, with Mutinex’s AI system, Hershey is now able run models in as little as three weeks.<p>In practice, that means faster iteration on how marketing spend is evaluated and adjusted, rather than waiting for lagging historical reads.<p>“Most companies don’t have an AI problem. They have a data readiness problem,” said Sarah Martinez, chief commercial officer, Tracer.<p></article excerpt><p>Instead of the headline, it sounds like they've hired an external company to clean up their ETL pipelines. That seems useful.<p>I'm going to doubt spooling up <massive LLM> with <appropriate system prompt> is going to be the thing that reduces their analysis time.</p>
]]></description><pubDate>Mon, 18 May 2026 13:29:55 +0000</pubDate><link>https://news.ycombinator.com/item?id=48179617</link><dc:creator>krapht</dc:creator><comments>https://news.ycombinator.com/item?id=48179617</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=48179617</guid></item><item><title><![CDATA[New comment by krapht in "Frontier AI has broken the open CTF format"]]></title><description><![CDATA[
<p>You can go your entire career without recursing, or using a tree data structure in its raw form (i.e. you only use it as part of a library)</p>
]]></description><pubDate>Sat, 16 May 2026 21:34:40 +0000</pubDate><link>https://news.ycombinator.com/item?id=48164020</link><dc:creator>krapht</dc:creator><comments>https://news.ycombinator.com/item?id=48164020</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=48164020</guid></item><item><title><![CDATA[New comment by krapht in "Sweden goes back to basics, swapping screens for books in the classroom"]]></title><description><![CDATA[
<p>which city is that? I want to move there with my children.<p>If you desperately want mediocre chromebook instruction for your children; where your classmates turn in ChatGPT essays as original work, feel free to move to Virginia.</p>
]]></description><pubDate>Thu, 02 Apr 2026 15:34:02 +0000</pubDate><link>https://news.ycombinator.com/item?id=47615860</link><dc:creator>krapht</dc:creator><comments>https://news.ycombinator.com/item?id=47615860</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=47615860</guid></item><item><title><![CDATA[New comment by krapht in "Ordered dithering with arbitrary or irregular colour palettes (2023)"]]></title><description><![CDATA[
<p>Lots of sensors these days will give you 10 or 12 bits of data per color channel. You may want ordered dithering when previewing on an 8 bit display.</p>
]]></description><pubDate>Mon, 23 Mar 2026 09:08:31 +0000</pubDate><link>https://news.ycombinator.com/item?id=47486954</link><dc:creator>krapht</dc:creator><comments>https://news.ycombinator.com/item?id=47486954</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=47486954</guid></item><item><title><![CDATA[New comment by krapht in "How HN: Ironkernel – Python expressions, Rust parallel"]]></title><description><![CDATA[
<p>Is the code actually good though? Not seeing any benchmarks vs numexpr, numba, or Jax</p>
]]></description><pubDate>Sat, 21 Mar 2026 16:44:29 +0000</pubDate><link>https://news.ycombinator.com/item?id=47468651</link><dc:creator>krapht</dc:creator><comments>https://news.ycombinator.com/item?id=47468651</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=47468651</guid></item><item><title><![CDATA[New comment by krapht in "WSL Manager"]]></title><description><![CDATA[
<p>Sublime Text? Sure, doesn't have the long tail of extensions, but surely most people don't need those. The biggest issue with ST being the fact that it costs money...</p>
]]></description><pubDate>Mon, 09 Mar 2026 03:44:53 +0000</pubDate><link>https://news.ycombinator.com/item?id=47304665</link><dc:creator>krapht</dc:creator><comments>https://news.ycombinator.com/item?id=47304665</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=47304665</guid></item><item><title><![CDATA[New comment by krapht in "A decade of Docker containers"]]></title><description><![CDATA[
<p>Paying customers use docker desktop.</p>
]]></description><pubDate>Sat, 07 Mar 2026 19:38:42 +0000</pubDate><link>https://news.ycombinator.com/item?id=47290772</link><dc:creator>krapht</dc:creator><comments>https://news.ycombinator.com/item?id=47290772</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=47290772</guid></item><item><title><![CDATA[New comment by krapht in "Pentagon formally labels Anthropic supply-chain risk"]]></title><description><![CDATA[
<p>Even if you don't like the current administration, the rank and file are still out there doing valuable work. The government is more than ICE; it also administers welfare, funds research, collects taxes, and distributes social security payments to the old and infirm.</p>
]]></description><pubDate>Thu, 05 Mar 2026 21:14:14 +0000</pubDate><link>https://news.ycombinator.com/item?id=47267376</link><dc:creator>krapht</dc:creator><comments>https://news.ycombinator.com/item?id=47267376</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=47267376</guid></item></channel></rss>