<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: jsemrau</title><link>https://news.ycombinator.com/user?id=jsemrau</link><description>Hacker News RSS</description><docs>https://hnrss.org/</docs><generator>hnrss v2.1.1</generator><lastBuildDate>Fri, 05 Jun 2026 00:58:21 +0000</lastBuildDate><atom:link href="https://hnrss.org/user?id=jsemrau" rel="self" type="application/rss+xml"></atom:link><item><title><![CDATA[New comment by jsemrau in "Anthropic surpasses OpenAI to become most valuable AI startup"]]></title><description><![CDATA[
<p>I did a pair programming comparison over 3 month on Codex 5.2 and Claude Sonnet and my subjective experience was that based on cost and rollbacks to a previous commit Claude is significantly better. Especially in VS Code Copilot. I wrote a long Substack post about it. I would share its but its in the paywalled archive by now.</p>
]]></description><pubDate>Sat, 30 May 2026 15:20:15 +0000</pubDate><link>https://news.ycombinator.com/item?id=48337191</link><dc:creator>jsemrau</dc:creator><comments>https://news.ycombinator.com/item?id=48337191</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=48337191</guid></item><item><title><![CDATA[New comment by jsemrau in "Expertise in the age of AI"]]></title><description><![CDATA[
<p>In my opinion, the bread example doesn't really work that well because it bridges into the physical domain which most cognitive systems don't have access to. That said, for grounding context and therefore creating truth having a version of a world model is very important (See Yan LeCun's work). My experience is that given the right world to operate in, an agent can indeed find flaws in recipes and fix it even though the agent has not been prompted explicitly to do it. This world, as far as I understand it now, is a combination of sequential (at which step am I in a process), conversational (what was talked about alread/ what had I done already), and context memory (what is the frame or reference/plane of existence).<p>Self-correcting agents are already here: <a href="https://jdsemrau.substack.com/p/hyperagents-and-self-correcting-systems" rel="nofollow">https://jdsemrau.substack.com/p/hyperagents-and-self-correct...</a></p>
]]></description><pubDate>Fri, 29 May 2026 19:35:46 +0000</pubDate><link>https://news.ycombinator.com/item?id=48328147</link><dc:creator>jsemrau</dc:creator><comments>https://news.ycombinator.com/item?id=48328147</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=48328147</guid></item><item><title><![CDATA[New comment by jsemrau in "I think Anthropic and OpenAI have found product-market fit"]]></title><description><![CDATA[
<p>Anthropic for sure. It's a useful professional product that I find many use-cases for in my professional and private life. OpenAI not so much.</p>
]]></description><pubDate>Thu, 28 May 2026 04:56:38 +0000</pubDate><link>https://news.ycombinator.com/item?id=48304699</link><dc:creator>jsemrau</dc:creator><comments>https://news.ycombinator.com/item?id=48304699</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=48304699</guid></item><item><title><![CDATA[New comment by jsemrau in "δ-mem: Efficient Online Memory for Large Language Models"]]></title><description><![CDATA[
<p>Thank you.</p>
]]></description><pubDate>Sun, 17 May 2026 07:15:49 +0000</pubDate><link>https://news.ycombinator.com/item?id=48166728</link><dc:creator>jsemrau</dc:creator><comments>https://news.ycombinator.com/item?id=48166728</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=48166728</guid></item><item><title><![CDATA[New comment by jsemrau in "δ-mem: Efficient Online Memory for Large Language Models"]]></title><description><![CDATA[
<p>That assumes one layer of memory. In my experience you need to have at least 4 layers of memory to work well. All of them have different requirements for retrieval. 
Everything that is in short-term memory (state of the app, current conversation, current workspace artefact) requires fast latency and precision. For example if you want to edit a segment in a financial analysis, a blog post, or a program you only want to edit this segment. RAG on a VectorDB is overkill in my opinion.</p>
]]></description><pubDate>Sun, 17 May 2026 00:04:58 +0000</pubDate><link>https://news.ycombinator.com/item?id=48164922</link><dc:creator>jsemrau</dc:creator><comments>https://news.ycombinator.com/item?id=48164922</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=48164922</guid></item><item><title><![CDATA[New comment by jsemrau in "δ-mem: Efficient Online Memory for Large Language Models"]]></title><description><![CDATA[
<p>I am currently working on deep context query which uses dynamically generated regex to pull only the relevant context blocks. By using lightweight RegEx pattern matching to detect semantic intent and filter structured context sections accordingly, you avoid the attention degradation that comes from stuffing semantically redundant information into the window<p><a href="https://jdsemrau.substack.com/p/tokenmaxxing-and-optimizing-context" rel="nofollow">https://jdsemrau.substack.com/p/tokenmaxxing-and-optimizing-...</a></p>
]]></description><pubDate>Sat, 16 May 2026 13:08:46 +0000</pubDate><link>https://news.ycombinator.com/item?id=48159949</link><dc:creator>jsemrau</dc:creator><comments>https://news.ycombinator.com/item?id=48159949</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=48159949</guid></item><item><title><![CDATA[New comment by jsemrau in "Codex is now in the ChatGPT mobile app"]]></title><description><![CDATA[
<p>I don't understand OpenAI's product strategy.</p>
]]></description><pubDate>Fri, 15 May 2026 01:21:10 +0000</pubDate><link>https://news.ycombinator.com/item?id=48143363</link><dc:creator>jsemrau</dc:creator><comments>https://news.ycombinator.com/item?id=48143363</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=48143363</guid></item><item><title><![CDATA[New comment by jsemrau in "Vibe coding and agentic engineering are getting closer than I'd like"]]></title><description><![CDATA[
<p>The same applies to banks and lending standards. 
In the end it is a function of governance and professional conduct.</p>
]]></description><pubDate>Wed, 06 May 2026 22:31:35 +0000</pubDate><link>https://news.ycombinator.com/item?id=48042764</link><dc:creator>jsemrau</dc:creator><comments>https://news.ycombinator.com/item?id=48042764</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=48042764</guid></item><item><title><![CDATA[New comment by jsemrau in "Uber wants to turn its drivers into a sensor grid for self-driving companies"]]></title><description><![CDATA[
<p>"You can’t be the CTO of Uber wanting to do AVs, and get the data collection requirement shockingly wrong."<p>Problem 1: Cost and privacy constrain limit data collection.<p>Problem 2: It makes not much sense to collect and store data that you already  have. Yet you don't know that when collecting if it is useful or not.<p>Problem 3: P2P in urban setting fails at edge cases which by definition are rare to collect.<p>All of these problems limit AV scaling.</p>
]]></description><pubDate>Sun, 03 May 2026 00:27:03 +0000</pubDate><link>https://news.ycombinator.com/item?id=47991948</link><dc:creator>jsemrau</dc:creator><comments>https://news.ycombinator.com/item?id=47991948</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=47991948</guid></item><item><title><![CDATA[New comment by jsemrau in "Regression: malware reminder on every read still causes subagent refusals"]]></title><description><![CDATA[
<p>When working with APIs it makes a lot of sense to filter only for relevant portions based on an intent-driven dynamic RegEx.</p>
]]></description><pubDate>Wed, 29 Apr 2026 01:57:27 +0000</pubDate><link>https://news.ycombinator.com/item?id=47943336</link><dc:creator>jsemrau</dc:creator><comments>https://news.ycombinator.com/item?id=47943336</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=47943336</guid></item><item><title><![CDATA[New comment by jsemrau in "John Ternus to become Apple CEO"]]></title><description><![CDATA[
<p>He is proof that LinkedIn doesn't matter.</p>
]]></description><pubDate>Tue, 21 Apr 2026 02:27:56 +0000</pubDate><link>https://news.ycombinator.com/item?id=47843820</link><dc:creator>jsemrau</dc:creator><comments>https://news.ycombinator.com/item?id=47843820</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=47843820</guid></item><item><title><![CDATA[New comment by jsemrau in "Codex for almost everything"]]></title><description><![CDATA[
<p>Codex is much worse than Anthropics model. My experience is that I burn 10x the tokens using Codex compared to Sonnet 4.6</p>
]]></description><pubDate>Thu, 16 Apr 2026 22:10:35 +0000</pubDate><link>https://news.ycombinator.com/item?id=47800184</link><dc:creator>jsemrau</dc:creator><comments>https://news.ycombinator.com/item?id=47800184</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=47800184</guid></item><item><title><![CDATA[New comment by jsemrau in "SWE-CI: Evaluating Agent Capabilities in Maintaining Codebases via CI"]]></title><description><![CDATA[
<p>It still fubars my code regularly at 11x the price. 
Github Copilot Agentic Mode + Sonnet 4.6 is stable and inexpensive.</p>
]]></description><pubDate>Mon, 09 Mar 2026 03:58:03 +0000</pubDate><link>https://news.ycombinator.com/item?id=47304737</link><dc:creator>jsemrau</dc:creator><comments>https://news.ycombinator.com/item?id=47304737</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=47304737</guid></item><item><title><![CDATA[New comment by jsemrau in "Ask HN: What Are You Working On? (March 2026)"]]></title><description><![CDATA[
<p>It's a personal project, but inspired by OpenClaw (which I find way overhyped), I am building an ambient intelligence layer for investment finance including a 3-tiered memory architecture, sensors (for environment scanning), skills, reasoning agents, and a new agentic UI concept only for that purpose.<p>I wrote about it here: <a href="https://jdsemrau.substack.com/p/pair-programming-superbill-with-codex" rel="nofollow">https://jdsemrau.substack.com/p/pair-programming-superbill-w...</a></p>
]]></description><pubDate>Mon, 09 Mar 2026 01:30:27 +0000</pubDate><link>https://news.ycombinator.com/item?id=47303770</link><dc:creator>jsemrau</dc:creator><comments>https://news.ycombinator.com/item?id=47303770</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=47303770</guid></item><item><title><![CDATA[New comment by jsemrau in "I ported Linux to the PS5 and turned it into a Steam Machine"]]></title><description><![CDATA[
<p>Most "hackers" now are build to sell (i.e., micro-entrepreneurship), not build for intellectual curiosity.</p>
]]></description><pubDate>Sun, 08 Mar 2026 23:09:23 +0000</pubDate><link>https://news.ycombinator.com/item?id=47302617</link><dc:creator>jsemrau</dc:creator><comments>https://news.ycombinator.com/item?id=47302617</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=47302617</guid></item><item><title><![CDATA[New comment by jsemrau in "I ported Linux to the PS5 and turned it into a Steam Machine"]]></title><description><![CDATA[
<p>Where did Fry's Electronics go? 
Where can I go an buy robot parts or even used tech parts?<p>Let's say I want to build a drone from scratch.</p>
]]></description><pubDate>Sun, 08 Mar 2026 23:07:00 +0000</pubDate><link>https://news.ycombinator.com/item?id=47302598</link><dc:creator>jsemrau</dc:creator><comments>https://news.ycombinator.com/item?id=47302598</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=47302598</guid></item><item><title><![CDATA[New comment by jsemrau in "SWE-CI: Evaluating Agent Capabilities in Maintaining Codebases via CI"]]></title><description><![CDATA[
<p>I reached the same conclusion. I tried using both for my personal investment ambient using agent-pair programming to build and agentic intelligence layer for stocks and the difference between the 2 models if astounding.</p>
]]></description><pubDate>Sun, 08 Mar 2026 13:44:41 +0000</pubDate><link>https://news.ycombinator.com/item?id=47297268</link><dc:creator>jsemrau</dc:creator><comments>https://news.ycombinator.com/item?id=47297268</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=47297268</guid></item><item><title><![CDATA[New comment by jsemrau in "I ported Linux to the PS5 and turned it into a Steam Machine"]]></title><description><![CDATA[
<p>We need to bring back this culture.</p>
]]></description><pubDate>Sun, 08 Mar 2026 13:32:24 +0000</pubDate><link>https://news.ycombinator.com/item?id=47297191</link><dc:creator>jsemrau</dc:creator><comments>https://news.ycombinator.com/item?id=47297191</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=47297191</guid></item><item><title><![CDATA[New comment by jsemrau in "GPT-5.4"]]></title><description><![CDATA[
<p>It was a different company back then. The Internet was still new-ish and not the multi-trillion dollar company it is now. I'd think expectations are different.</p>
]]></description><pubDate>Fri, 06 Mar 2026 02:42:18 +0000</pubDate><link>https://news.ycombinator.com/item?id=47270174</link><dc:creator>jsemrau</dc:creator><comments>https://news.ycombinator.com/item?id=47270174</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=47270174</guid></item><item><title><![CDATA[New comment by jsemrau in "Microsoft announces new "mini PCs" for Windows 365"]]></title><description><![CDATA[
<p>1997 Sun rays.</p>
]]></description><pubDate>Sun, 01 Mar 2026 04:34:20 +0000</pubDate><link>https://news.ycombinator.com/item?id=47203750</link><dc:creator>jsemrau</dc:creator><comments>https://news.ycombinator.com/item?id=47203750</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=47203750</guid></item></channel></rss>