<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: jimmytucson</title><link>https://news.ycombinator.com/user?id=jimmytucson</link><description>Hacker News RSS</description><docs>https://hnrss.org/</docs><generator>hnrss v2.1.1</generator><lastBuildDate>Thu, 30 Apr 2026 12:59:13 +0000</lastBuildDate><atom:link href="https://hnrss.org/user?id=jimmytucson" rel="self" type="application/rss+xml"></atom:link><item><title><![CDATA[New comment by jimmytucson in "Goodbye to Sora"]]></title><description><![CDATA[
<p>Pretty much mirrors my experience using GPT to generate images creatively. I tried to generate an image to accompany a Robert frost poem and it made something... plausibly related. But not what I was describing. I spent the next 90% of the time making it 10% closer to what I wanted but it never got all the way there.<p>I’ve given it different levels of open-endednes, give this flow chart an aesthetic like this mechanical keyboard, or generate an SVG of this graphic from a 70s slide show, but it never looks quite like what I have in mind.<p>In the end, I think you only use this stuff to generate images if you’re prepared to accept whatever comes out on approximately the first try.</p>
]]></description><pubDate>Tue, 24 Mar 2026 22:48:37 +0000</pubDate><link>https://news.ycombinator.com/item?id=47510636</link><dc:creator>jimmytucson</dc:creator><comments>https://news.ycombinator.com/item?id=47510636</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=47510636</guid></item><item><title><![CDATA[New comment by jimmytucson in "Claude March 2026 usage promotion"]]></title><description><![CDATA[
<p>They have this. It’s called batch pricing and it’s 50% off.</p>
]]></description><pubDate>Sat, 14 Mar 2026 22:06:55 +0000</pubDate><link>https://news.ycombinator.com/item?id=47381763</link><dc:creator>jimmytucson</dc:creator><comments>https://news.ycombinator.com/item?id=47381763</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=47381763</guid></item><item><title><![CDATA[New comment by jimmytucson in "The Cheese and the Worms"]]></title><description><![CDATA[
<p>> I have said that, in my opinion, all was chaos, that is, earth, air, water, and fire were mixed together; and out of that bulk a mass formed – just as cheese is made out of milk – and worms appeared in it, and these were the angels. The most holy majesty decreed that these should be God and the angels, and among that number of angels there was also God, he too having been created out of that mass at the same time, and he was named lord with four captains, Lucifer, Michael, Gabriel, and Raphael. That Lucifer sought to make himself lord equal to the king, who was the majesty of God, and for this arrogance God ordered him driven out of heaven with all his host and his company; and this God later created Adam and Eve and people in great number to take the places of the angels who had been expelled. And as this multitude did not follow God's commandments, he sent his Son, whom the Jews seized, and he was crucified.<p>Even though this sounded nuts to people in Menocchio’s time, it’s not completely implausible. The whole thing about order arising from undifferentiated chaos is pretty Greek, as is the idea of life arising spontaneously from decaying matter (Aristotle believed that). God arising from some preexisting substance is kind of radical but I’m sure there were precursors to that too. The rest of it is almost stock standard Christian mythology.</p>
]]></description><pubDate>Fri, 13 Mar 2026 01:48:00 +0000</pubDate><link>https://news.ycombinator.com/item?id=47359761</link><dc:creator>jimmytucson</dc:creator><comments>https://news.ycombinator.com/item?id=47359761</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=47359761</guid></item><item><title><![CDATA[New comment by jimmytucson in "The rise of eyes began with just one"]]></title><description><![CDATA[
<p>That “cyclopean” eye is described as a patch of light-sensitive cells.</p>
]]></description><pubDate>Tue, 24 Feb 2026 02:41:39 +0000</pubDate><link>https://news.ycombinator.com/item?id=47132200</link><dc:creator>jimmytucson</dc:creator><comments>https://news.ycombinator.com/item?id=47132200</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=47132200</guid></item><item><title><![CDATA[New comment by jimmytucson in "TimeCapsuleLLM: LLM trained only on data from 1800-1875"]]></title><description><![CDATA[
<p>Fascinating idea.  There was another "time-locked" LLM project that popped up on HN recently[1].  Their model output is really polished but the team is trying to figure out how to avoid abuse and misrepresentation of their goals.  We think it would be cool to talk to someone from 100+ years ago but haven't seriously considered the many ways in which it would be uncool. Interesting times!<p>[1] <a href="https://news.ycombinator.com/item?id=46319826">https://news.ycombinator.com/item?id=46319826</a></p>
]]></description><pubDate>Mon, 12 Jan 2026 17:24:53 +0000</pubDate><link>https://news.ycombinator.com/item?id=46591501</link><dc:creator>jimmytucson</dc:creator><comments>https://news.ycombinator.com/item?id=46591501</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=46591501</guid></item><item><title><![CDATA[New comment by jimmytucson in "An Honest Review of Go (2025)"]]></title><description><![CDATA[
<p>You can definitely protect from reassignment in those languages (e.g. `final` in Java) but they don't completely prevent you from changing the underlying data.  Rust would be one that comes to mind that has true immutability.  I guess the Go maintainers just didn't want to go down that road, which I get.</p>
]]></description><pubDate>Thu, 08 Jan 2026 19:27:11 +0000</pubDate><link>https://news.ycombinator.com/item?id=46545305</link><dc:creator>jimmytucson</dc:creator><comments>https://news.ycombinator.com/item?id=46545305</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=46545305</guid></item><item><title><![CDATA[New comment by jimmytucson in "An Honest Review of Go (2025)"]]></title><description><![CDATA[
<p>Not shilling for Go here but there are a few misconceptions in this blog post. In addition to the others mentioned:<p>> All of these examples involve assigning to a constant a value known at compile time but none of them will work<p>Maps are not known at compile time.  Hash functions are randomized based on a seed only known at execution time. The hashed value of "HELLO" is actually different each time the program runs. Even if the hash function weren't random, the runtime has to allocate buckets for map values <i>on the heap</i>, which involves calling the OS to get memory addresses for those buckets, etc.<p>In Go, `const` means "the compiler can completely evaluate this expression and store the final bytes in the executable," which has the effect of making them non-reassignable, but protection from reassignment is not an actual feature of the language the way it is in Java and C++ (goes back to the maintainers wanting to keep it simple).</p>
]]></description><pubDate>Thu, 08 Jan 2026 17:20:09 +0000</pubDate><link>https://news.ycombinator.com/item?id=46543623</link><dc:creator>jimmytucson</dc:creator><comments>https://news.ycombinator.com/item?id=46543623</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=46543623</guid></item><item><title><![CDATA[New comment by jimmytucson in "C Is Best (2025)"]]></title><description><![CDATA[
<p>Hmm, so in a language that does automatic bounds checking, the compiler might translate a line of source code like:<p><pre><code>    let val = arr[i]
</code></pre>
to assembly code like:<p><pre><code>    cmp     rdx, rsi        ; Compare i (rdx) with length (rsi)
    jae     .Lpanic_label   ; Jump if i >= length
    ; later...
    .Lpanic_label:
    call    core::panicking::panic_bounds_check

</code></pre>
Are they saying with "correct code" the line of source code won't be covered?  Because the assembly instruction to call panic isn't ever reached?</p>
]]></description><pubDate>Tue, 06 Jan 2026 22:09:28 +0000</pubDate><link>https://news.ycombinator.com/item?id=46519548</link><dc:creator>jimmytucson</dc:creator><comments>https://news.ycombinator.com/item?id=46519548</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=46519548</guid></item><item><title><![CDATA[New comment by jimmytucson in "C Is Best (2025)"]]></title><description><![CDATA[
<p>> Safe languages insert additional machine branches to do things like verify that array accesses are in-bounds. In correct code, those branches are never taken. That means that the machine code cannot be 100% branch tested, which is an important component of SQLite's quality strategy.<p>Doesn't the language compiler write the code that checks if the array access is in-bounds?  Why would you need to test the compiler's code?</p>
]]></description><pubDate>Tue, 06 Jan 2026 17:25:57 +0000</pubDate><link>https://news.ycombinator.com/item?id=46515385</link><dc:creator>jimmytucson</dc:creator><comments>https://news.ycombinator.com/item?id=46515385</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=46515385</guid></item><item><title><![CDATA[New comment by jimmytucson in "Open-source Zig book"]]></title><description><![CDATA[
<p>> [Learning Zig] is about fundamentally changing how you think about software.<p>Learning LISP, Fortran, APL, Perl, or really any language that is different from what you’re used to, will also do this for you.</p>
]]></description><pubDate>Mon, 17 Nov 2025 03:57:18 +0000</pubDate><link>https://news.ycombinator.com/item?id=45950733</link><dc:creator>jimmytucson</dc:creator><comments>https://news.ycombinator.com/item?id=45950733</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=45950733</guid></item><item><title><![CDATA[New comment by jimmytucson in "Building the heap: racking 30 petabytes of hard drives for pretraining"]]></title><description><![CDATA[
<p>Just wanted to say, thanks for doing this! Now the old rant...<p>I started my career when on-prem was the norm and remember so much trouble. When you have long-lived hardware, eventually, no matter how hard you try, you just start to treat it as a pet and state naturally accumulates. Then, as the hardware starts to be not good enough, you need to upgrade. There's an internal team that presents the "commodity" interface, so you have to pick out your new hardware from their list and get the cost approved (it's a lot harder to just spend a little more and get a little more). Then your projects are delayed by them racking the new hardware and you properly "un-petting" your pets so they can respawn on the new devices, etc.<p>Anyways, when cloud came along, I was like, yeah we're switching and never going back.  Buuut, come to find out that's part of the master plan: it's a no-brainer good deal until you and everyone in your org/company/industry forgets HTF to rack their own hardware, and then it starts to go from no-brainer to brainer.  And basically unless you start to pull back and rebuild that muscle, it will go from brainer to no-brainer <i>bad</i> deal. So thanks for building this muscle!</p>
]]></description><pubDate>Wed, 01 Oct 2025 17:04:27 +0000</pubDate><link>https://news.ycombinator.com/item?id=45440160</link><dc:creator>jimmytucson</dc:creator><comments>https://news.ycombinator.com/item?id=45440160</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=45440160</guid></item><item><title><![CDATA[New comment by jimmytucson in "Show HN: PageIndex – Vectorless RAG"]]></title><description><![CDATA[
<p>It is just as "vibe-ish" as vector search and notably does require chunking (document chunks are fed to the indexer to build the table of contents). That said, I don't find vector search any less "vibey".  While "mathematical similarity" is a structured operation, the "conversion to high-dimensional vectors" part is predicated on the encoder, which can be trained towards any objective.<p><pre><code>    > scaling will become problematic as the doc structure approaches the context limit of the LLM doing the retrieval
</code></pre>
IIUC, retrieval is based on traversing a tree structure, so only the root nodes have to fit in the context window. I find that kinda cool about this approach.<p>But yes, still "vibe retrieval".</p>
]]></description><pubDate>Fri, 29 Aug 2025 15:57:00 +0000</pubDate><link>https://news.ycombinator.com/item?id=45065752</link><dc:creator>jimmytucson</dc:creator><comments>https://news.ycombinator.com/item?id=45065752</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=45065752</guid></item><item><title><![CDATA[New comment by jimmytucson in "Every industry is an overcrowded airport lounge now"]]></title><description><![CDATA[
<p>That’s because every business is “scaled” to the point that the edge employees —- ie the people who interact with paying customers —- don’t own anything, and are 12 levels of management away from anyone who does.<p>My grandparents owned a grocery store. Their name was on the sign. If you brought home spoiled meat, that was their name and you as a member of their community that were put out.<p>When my mom brings home spoiled meat from Stop & Shop, she goes back there not just to exchange it, but to complain to someone about how it messed up her barbecue plans, etc. And I’m like seriously, why would anyone working at Stop & Shop give a rat’s ass about your family gathering? Stop & Shop is owned by a Dutch multinational “food retail” company.<p>But that’s not the capitalism she grew up with. She actually thinks capitalism is great because it allowed her parents to come over on a boat as teenagers and make lives for themselves, and have extra to send back home. But she hates it when she calls her cable company and ends up chatting with a girl in Singapore. Go figure.</p>
]]></description><pubDate>Fri, 29 Aug 2025 03:13:10 +0000</pubDate><link>https://news.ycombinator.com/item?id=45059728</link><dc:creator>jimmytucson</dc:creator><comments>https://news.ycombinator.com/item?id=45059728</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=45059728</guid></item><item><title><![CDATA[New comment by jimmytucson in "CEO pay and stock buybacks have soared at the largest low-wage corporations"]]></title><description><![CDATA[
<p>The company with the lowest median pay, Aptiv, I had to look up, and what a darling of a company this is:<p><pre><code>    - guilty of systemic accounting fraud from 1999-2004
    - manufactured a (allegedly known-to-be) faulty ignition switch that led to the deaths of 124 people and injured 275 others
    - currently forcing 20,000 former employees to fight a decades-long legal battle for their earned pension benefits
    - gleefully dumps massive quantities of carcinogens and poisons into the environment, including: lead compounds, chromium compounds, sulfuric and hydrochloride acid (lol), and glycol ethers
</code></pre>
Simply a masterclass in corporate irresponsibility, exactly why their CEO is so well comped.<p>Citations here: <a href="https://en.m.wikipedia.org/wiki/Aptiv" rel="nofollow">https://en.m.wikipedia.org/wiki/Aptiv</a></p>
]]></description><pubDate>Fri, 22 Aug 2025 02:24:59 +0000</pubDate><link>https://news.ycombinator.com/item?id=44980432</link><dc:creator>jimmytucson</dc:creator><comments>https://news.ycombinator.com/item?id=44980432</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=44980432</guid></item><item><title><![CDATA[New comment by jimmytucson in "Show HN: Vectorless RAG"]]></title><description><![CDATA[
<p>> That’s why we use a tree structure rather than just a flat list of sections. This is what makes it different from traditional RAG<p>Ah ok, that’s a key piece I was missing. That’s really cool, thanks!</p>
]]></description><pubDate>Thu, 21 Aug 2025 17:03:18 +0000</pubDate><link>https://news.ycombinator.com/item?id=44975213</link><dc:creator>jimmytucson</dc:creator><comments>https://news.ycombinator.com/item?id=44975213</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=44975213</guid></item><item><title><![CDATA[New comment by jimmytucson in "Show HN: Vectorless RAG"]]></title><description><![CDATA[
<p>So if I understand this correctly, this works on a single large document whose size exceeds what you can or want to put into a single context frame for answering a question? It first "indexes" the document by feeding successive "proto-chunks" to an LLM, along with an accumulator, which is like a running table of contents into the document with "sections" that the indexer LLM decides on and summarizes, until the table of contents is complete. (What we're calling "sections" here - these are still "chunks", they're just not a fixed size and are decided on by the indexer at build time?)<p>Then for the retrieval stage, it presents the table of contents to a "retriever" LLM, which decides which sections are relevant to the question based on the summaries the indexer LLM created. Then for the answer generation stage, it just presents those relevant sections along with the question.<p>That's pretty clever - does it work with a corpus of documents as well, or just a single large document? Does the "indexer" know the question ahead of time, or is the creation of sections and section summarization supposed to be question-agnostic?  What if your table of contents gets too big?  Seems like then it just becomes normal RAG, where you have to store the summaries and document-chunk pointers in some vector or lexical database?</p>
]]></description><pubDate>Wed, 20 Aug 2025 20:10:18 +0000</pubDate><link>https://news.ycombinator.com/item?id=44965901</link><dc:creator>jimmytucson</dc:creator><comments>https://news.ycombinator.com/item?id=44965901</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=44965901</guid></item><item><title><![CDATA[New comment by jimmytucson in "19% of California houses are owned by investors"]]></title><description><![CDATA[
<p>To save you a click, 19% is actually <i>not</i> a lot (I thought it was):<p>> 19% of California houses were owned by investors, ranking No. 36 among the states and just below the 20% national norm.<p>States with the highest share of investor-owned houses:<p>> Hawaii at 40%, Alaska at 35%, Vermont at 31%, West Virginia at 30%, and Wyoming at 30%.<p>States with the lowest are all in the Mid-Atlantic and lower New England:<p>> Connecticut at 10%, Rhode Island and  Massachusetts at 12%, and Delaware at 13%.<p>Why so low in California (again, I'm baffled that this is "low")?<p>> the sky-high price tag for single-family homes, the third-highest nationally at $866,100</p>
]]></description><pubDate>Wed, 06 Aug 2025 19:49:58 +0000</pubDate><link>https://news.ycombinator.com/item?id=44816839</link><dc:creator>jimmytucson</dc:creator><comments>https://news.ycombinator.com/item?id=44816839</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=44816839</guid></item><item><title><![CDATA[New comment by jimmytucson in "Disney's ESPN to acquire NFL media assets in major deal"]]></title><description><![CDATA[
<p>Pretty amazing this all started as MLB Advanced Media more than 20 years ago streaming baseball games on the internet before YouTube existed! When I joined, they were streaming NHL, MLS, WWE, HBO, and many others. Then they spun off and sold to Disney and became Disney+. I wonder if any lines of code from those very early days still get executed.<p>Regardless, this is terrible for sports fans. Disney will chop this up into “baskets” with one watchable thing padded with lots of other unwatchable-to-mildly-interesting stuff (college hockey from a single camera angle anyone?). Then, when you want to watch some event, you’ll open your ESPN Premium +- whatever app, get excited when you see the event on the home screen, only to be upsold to start your free trial of ESPN NFL++ Hulu Fans Only Bundle. The only solution is to boycott the whole damn thing, which is what I am doing, and I love having the extra cash for tickets to local minor league games, etc.</p>
]]></description><pubDate>Wed, 06 Aug 2025 01:16:00 +0000</pubDate><link>https://news.ycombinator.com/item?id=44806567</link><dc:creator>jimmytucson</dc:creator><comments>https://news.ycombinator.com/item?id=44806567</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=44806567</guid></item><item><title><![CDATA[New comment by jimmytucson in "Job-seekers are dodging AI interviewers"]]></title><description><![CDATA[
<p>Why not do it before you need a job?  While you're comfortable, submit your application for open roles and reject the AI interviewer.</p>
]]></description><pubDate>Mon, 04 Aug 2025 17:44:30 +0000</pubDate><link>https://news.ycombinator.com/item?id=44789117</link><dc:creator>jimmytucson</dc:creator><comments>https://news.ycombinator.com/item?id=44789117</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=44789117</guid></item><item><title><![CDATA[New comment by jimmytucson in "Gemini North telescope discovers long-predicted stellar companion of Betelgeuse"]]></title><description><![CDATA[
<p>The system’s brightness decreases when the companion star swings around behind Betelgeuse. It also dips when Betelgeuse goes behind the companion star but much less so because Betelgeuse is so much larger.</p>
]]></description><pubDate>Wed, 23 Jul 2025 03:35:03 +0000</pubDate><link>https://news.ycombinator.com/item?id=44655498</link><dc:creator>jimmytucson</dc:creator><comments>https://news.ycombinator.com/item?id=44655498</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=44655498</guid></item></channel></rss>