<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: jjcm</title><link>https://news.ycombinator.com/user?id=jjcm</link><description>Hacker News RSS</description><docs>https://hnrss.org/</docs><generator>hnrss v2.1.1</generator><lastBuildDate>Wed, 18 Nov 2026 03:54:11 +0000</lastBuildDate><atom:link href="https://hnrss.org/user?id=jjcm" rel="self" type="application/rss+xml"></atom:link><item><title><![CDATA[New comment by jjcm in "Charles Schwab website goes down"]]></title><description><![CDATA[
<p>If I had to choose between:<p>- stable always-on service that never goes down<p>And<p>- down 1% of the time, but FDIC insured store of assets<p>I know which of the two I’d choose amidst an unparalleled level of security risk</p>
]]></description><pubDate>Tue, 17 Nov 2026 17:27:33 +0000</pubDate><link>https://news.ycombinator.com/item?id=50534920</link><dc:creator>jjcm</dc:creator><comments>https://news.ycombinator.com/item?id=50534920</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50534920</guid></item><item><title><![CDATA[New comment by jjcm in "Gemini 4 Hydrogen"]]></title><description><![CDATA[
<p>Indeed! Though I will say my ~5 years working at Figma did help in that respect.</p>
]]></description><pubDate>Sun, 15 Nov 2026 20:51:22 +0000</pubDate><link>https://news.ycombinator.com/item?id=50510603</link><dc:creator>jjcm</dc:creator><comments>https://news.ycombinator.com/item?id=50510603</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50510603</guid></item><item><title><![CDATA[New comment by jjcm in "Parents pay $75k for Alpha's AI school. Half of its students missed math target"]]></title><description><![CDATA[
<p>> the pitch is that students will “learn twice as much as their peers in two hours of academics a day”<p>Given a standard school day has 6 hours of learning, what they're actually claiming is that their product will teach kids 6x as effectively as normal schooling.<p>I do actually thing that AI-driven lessons are the future (I'd love to see a proper Young Lady's Illustrated Primer), but in my opinion the approach here is fundamentally flawed. The goal shouldn't be "less time spent learning", it should be to maximize the impact of the time spent.<p>The time spent learning is education in its own, and minimizing it to two hours a day will only teach kids to search for shortcuts and reduce their resiliance. You can certainly go too far with this (imo a Kumon-style 8am-8pm 12 hour day of learning overloads kids and doesn't give them time for creative thinking), but I do think it's important to teach kids that dedicated time spent on a problem is valuable in its own right.<p>Alpha's take is that this time is better spent on "real world" exercises and scenarios. From their site:<p>> Whether it’s solving a 45-second pit stop challenge or exploring the breadth of global art styles, our workshops teach students how to navigate the real world with confidence.<p>There is value in this iff the real world exercises are applications of the things they're learning. I was home schooled, and one of the most valuable things my parents did was give me things that were applications of what I was working on in my coursework. When I was learning geometry, one of my after school activities was calculating how much wood was needed for a deck that had a sine-wave curve and was made of 2x6 boards with a quarter inch gap. Once I did the calcs, we went to home depot, bought the wood, built the deck, and when I was a few boards short we talked through what I did wrong. It was a perfect way to reinforce how the geometry calculations could be applied. Alpha seems to focus more of their workshops on resiliance and leadership - ie can they bike all the way across the golden gate bridge, or can they do a ninja obstacle course (<a href="https://www.youtube.com/watch?v=5Vo16UaJ26o&t=197s" rel="nofollow">https://www.youtube.com/watch?v=5Vo16UaJ26o&t=197s</a>). I don't see these activities reinforcing what the kids are learning that day, so I'm not surprised they're falling short of the claimed targets.</p>
]]></description><pubDate>Sun, 15 Nov 2026 20:45:53 +0000</pubDate><link>https://news.ycombinator.com/item?id=50510544</link><dc:creator>jjcm</dc:creator><comments>https://news.ycombinator.com/item?id=50510544</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50510544</guid></item><item><title><![CDATA[New comment by jjcm in "Gemini 4 Hydrogen"]]></title><description><![CDATA[
<p>I'm very impressed with it as a model for image->html builds. It's the first non-anthropic low-param model that successfully captures complex flows.<p>Comparisons:<p>Hydrogen: <a href="https://html.non.io/lcars-gemini-4-hydrogen" rel="nofollow">https://html.non.io/lcars-gemini-4-hydrogen</a><p>Sonnet 5.5: <a href="https://html.non.io/lcars-sonnet-5.5/" rel="nofollow">https://html.non.io/lcars-sonnet-5.5/</a><p>Original designs: <a href="https://diffui.ai/app/canvas/5093e689-1e74-4f26-b632-2a4500f0da97?share=65ac3348-895b-4264-824e-b5f343040009" rel="nofollow">https://diffui.ai/app/canvas/5093e689-1e74-4f26-b632-2a4500f...</a><p>IMO it sits between sonnet and opus in terms of quality, and is equivalent with gpt sol (for this particular benchmark). Google did a great job with this.</p>
]]></description><pubDate>Thu, 12 Nov 2026 21:57:48 +0000</pubDate><link>https://news.ycombinator.com/item?id=50474893</link><dc:creator>jjcm</dc:creator><comments>https://news.ycombinator.com/item?id=50474893</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50474893</guid></item><item><title><![CDATA[New comment by jjcm in "Nintendo Takes Down Emulators and Firmware in Massive GitHub DMCA Sweep"]]></title><description><![CDATA[
<p>Low complexity, but to test fable 5.2 I gave it this instagram reel of a
Kikis delivery service inspired seedance video: <a href="https://www.instagram.com/reel/DXKjMSIkv97/?mdxt=OHBwaW5yamVkaDU1" rel="nofollow">https://www.instagram.com/reel/DXKjMSIkv97/?mdxt=OHBwaW5yamV...</a><p>And told it to make a game, and it made: <a href="https://petalwake.non.io" rel="nofollow">https://petalwake.non.io</a><p>Vid is here if you prefer that: <a href="https://x.com/pwnies/status/2119987174884143348?s=46&t=bwJTI_vAYdhYmGatsP4LjQ" rel="nofollow">https://x.com/pwnies/status/2119987174884143348?s=46&t=bwJTI...</a></p>
]]></description><pubDate>Thu, 12 Nov 2026 18:00:11 +0000</pubDate><link>https://news.ycombinator.com/item?id=50471408</link><dc:creator>jjcm</dc:creator><comments>https://news.ycombinator.com/item?id=50471408</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50471408</guid></item><item><title><![CDATA[New comment by jjcm in "Why are we still working this hard?"]]></title><description><![CDATA[
<p>Because capitalism is a competitive environment, and improvements in efficiency <i>increase</i> the opportunity cost of not working harder.<p>We aren't competing with workers from 1973, we're competing with other workers leveraging the same accelerated efficiency.</p>
]]></description><pubDate>Wed, 11 Nov 2026 20:39:01 +0000</pubDate><link>https://news.ycombinator.com/item?id=50458352</link><dc:creator>jjcm</dc:creator><comments>https://news.ycombinator.com/item?id=50458352</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50458352</guid></item><item><title><![CDATA[New comment by jjcm in "Claude Fable 5.2 and Mythos 5.2"]]></title><description><![CDATA[
<p>Re the twitter trend prompt, one thing I added to mine that improved the outcome was this blurb:<p><pre><code>    Look at the code pathways for anything you find. Look through the git history for each one. Understand why code was written the way it was. Revisit each hypothesis and give me a full plan again after you're done.</code></pre></p>
]]></description><pubDate>Tue, 10 Nov 2026 01:22:30 +0000</pubDate><link>https://news.ycombinator.com/item?id=50434867</link><dc:creator>jjcm</dc:creator><comments>https://news.ycombinator.com/item?id=50434867</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50434867</guid></item><item><title><![CDATA[New comment by jjcm in "Claude Fable 5.2 and Mythos 5.2"]]></title><description><![CDATA[
<p>Heavy agree. For high level architecture work, refactors, or any other high int problems, I use Fable and Astra. Whichever one starts the work, the other reviews the work / makes suggestions.<p>There's a trend on twitter going around telling you to paste the following into your agent, and it's the kind of work I go straight to fable+astra for:<p><pre><code>    Find all of the implicit state machines in my app. Make recommendations for changes that will elevate all three: the user experience (UX), developer experience (DX), and agent experience (AX) always, without breaking anything.</code></pre></p>
]]></description><pubDate>Tue, 10 Nov 2026 01:20:36 +0000</pubDate><link>https://news.ycombinator.com/item?id=50434856</link><dc:creator>jjcm</dc:creator><comments>https://news.ycombinator.com/item?id=50434856</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50434856</guid></item><item><title><![CDATA[New comment by jjcm in "Claude Fable 5.2 scores 65 on Artificial Analysis, highest so far"]]></title><description><![CDATA[
<p>> I think it will end up bad for them<p>There will always be a market for the most intelligent model. Keep in mind as well that something like Fable doesn't have to be cost effective on its own. Fable is Anthropic's way of building datasets for smaller more dense models, which is viable on its own as a method of success. These large labs are constrained by their dataset more than anything else, and these expensive-but-intelligent models create some of the best data out there.</p>
]]></description><pubDate>Tue, 10 Nov 2026 00:19:47 +0000</pubDate><link>https://news.ycombinator.com/item?id=50434385</link><dc:creator>jjcm</dc:creator><comments>https://news.ycombinator.com/item?id=50434385</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50434385</guid></item><item><title><![CDATA[New comment by jjcm in "Claude Fable 5.2 and Mythos 5.2"]]></title><description><![CDATA[
<p>This is actually something I'm working on as part of a more formalized version of this benchmark, both as a tool to assist LLMs in doing this (especially for ones with a more poor vision model), but also to properly measure how well they've done.<p>Still in progress, but early demo showing difference zones: <a href="https://non.io/video/fidelity-lens.mp4" rel="nofollow">https://non.io/video/fidelity-lens.mp4</a></p>
]]></description><pubDate>Mon, 09 Nov 2026 20:51:14 +0000</pubDate><link>https://news.ycombinator.com/item?id=50431928</link><dc:creator>jjcm</dc:creator><comments>https://news.ycombinator.com/item?id=50431928</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50431928</guid></item><item><title><![CDATA[New comment by jjcm in "Claude Fable 5.2 and Mythos 5.2"]]></title><description><![CDATA[
<p>I've been surprised at how long it's taken for this particular test to become fully saturated. So far there hasn't been a model that's able to perfectly one shot it.</p>
]]></description><pubDate>Mon, 09 Nov 2026 20:47:48 +0000</pubDate><link>https://news.ycombinator.com/item?id=50431874</link><dc:creator>jjcm</dc:creator><comments>https://news.ycombinator.com/item?id=50431874</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50431874</guid></item><item><title><![CDATA[New comment by jjcm in "Claude Fable 5.2 and Mythos 5.2"]]></title><description><![CDATA[
<p>That's me versus fable. I only told fable to build out the first page rather than the full app that's in the designs (was approaching the limit of my claude plan).</p>
]]></description><pubDate>Mon, 09 Nov 2026 20:10:44 +0000</pubDate><link>https://news.ycombinator.com/item?id=50431231</link><dc:creator>jjcm</dc:creator><comments>https://news.ycombinator.com/item?id=50431231</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50431231</guid></item><item><title><![CDATA[New comment by jjcm in "Claude Fable 5.2 and Mythos 5.2"]]></title><description><![CDATA[
<p>Image to html comparisons of Fable 5.2 and Opus 5.5 (the previous SOTA for this test):<p>Fable 5.2: <a href="https://html.non.io/lcars-fable-5.2/" rel="nofollow">https://html.non.io/lcars-fable-5.2/</a><p>Opus 5.5: <a href="https://html.non.io/lcars-opus-5.5/" rel="nofollow">https://html.non.io/lcars-opus-5.5/</a><p>Source designs: <a href="https://diffui.ai/app/canvas/5093e689-1e74-4f26-b632-2a4500f0da97?share=65ac3348-895b-4264-824e-b5f343040009" rel="nofollow">https://diffui.ai/app/canvas/5093e689-1e74-4f26-b632-2a4500f...</a><p>Thoughts:<p>The best implementation wrt being as true as possible to the source design. It nails the flowing svg shapes. Icons are more accurate to the design - note the difference between the starship svg in the "Legendary platforms" card between Opus 5.5 and Fable 5.2. The 3js planets outside the windows are also really well done, especially the bottom one which has a nice background glow to it.<p>It isn't without issues though - the flicker effect it added to the S shaped svg that weaves through the page is distracting in my opinion, and I actually thing the motion animation in the Opus 5.5 one is better overall. Still, it nails the implementation of this page, but it's no surprise considering how good Opus 5.5 was at this.<p>Disclaimer / reminder: the objective of this test isn't to have a well designed webpage, it's a synthetic test that pushes the limits of LLM's vision capabilities. Traditionally LLMs have issues with smooth curves, svg animations, and non-boxy layouts, which this design pushes the limits of.</p>
]]></description><pubDate>Mon, 09 Nov 2026 19:54:19 +0000</pubDate><link>https://news.ycombinator.com/item?id=50430962</link><dc:creator>jjcm</dc:creator><comments>https://news.ycombinator.com/item?id=50430962</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50430962</guid></item><item><title><![CDATA[New comment by jjcm in "Ask HN: What are you working on? (November 2026)"]]></title><description><![CDATA[
<p>I'm working on a better way to design longer scrolling pages with multiple sections, while preserving the visual look/feel and drastically speeding up the experience.<p>Here's a demo: <a href="https://non.io/video/linen-demo.mp4" rel="nofollow">https://non.io/video/linen-demo.mp4</a><p>As a disclaimer I cut 32 seconds of generation time out of that video, but still I'm quite happy with the speed. Not launched yet, but it's been fun to work on / wanted to show off progress :)</p>
]]></description><pubDate>Sun, 08 Nov 2026 22:47:57 +0000</pubDate><link>https://news.ycombinator.com/item?id=50418326</link><dc:creator>jjcm</dc:creator><comments>https://news.ycombinator.com/item?id=50418326</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50418326</guid></item><item><title><![CDATA[New comment by jjcm in "The Ubuntu 26.04 Rust sudo is not compatible with standard sudo"]]></title><description><![CDATA[
<p>doh - im an idiot. I read that as "from debian", assuming they were referring to the wider debian suite. TY.</p>
]]></description><pubDate>Fri, 06 Nov 2026 23:05:34 +0000</pubDate><link>https://news.ycombinator.com/item?id=50399717</link><dc:creator>jjcm</dc:creator><comments>https://news.ycombinator.com/item?id=50399717</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50399717</guid></item><item><title><![CDATA[New comment by jjcm in "The Ubuntu 26.04 Rust sudo is not compatible with standard sudo"]]></title><description><![CDATA[
<p>As someone who's been using ubuntu since warty warthog, what did you end up switching to / what are other people using these days?</p>
]]></description><pubDate>Fri, 06 Nov 2026 22:58:43 +0000</pubDate><link>https://news.ycombinator.com/item?id=50399638</link><dc:creator>jjcm</dc:creator><comments>https://news.ycombinator.com/item?id=50399638</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50399638</guid></item><item><title><![CDATA[New comment by jjcm in "Images 2.6"]]></title><description><![CDATA[
<p>IMO a lot of that is steering from the chatgpt harness. If you ask chatgpt for "a poster showing my latest burger specials", it really steers it towards the same look as everything else.<p>Try it via the API with your own LLM for stylistic steering. You'll find that chatgpt actually obfuscates how good the model actually is. You can get some serious variety out of it.<p>If you want some examples, halfway down the page I show comparisons of the same bookstore website in different styles - all of these were from gpt image: <a href="https://diffui.ai/for/bootstrappers" rel="nofollow">https://diffui.ai/for/bootstrappers</a> (disclaimer that this is a product I work on blahblahblah capitalism)</p>
]]></description><pubDate>Fri, 06 Nov 2026 22:57:34 +0000</pubDate><link>https://news.ycombinator.com/item?id=50399631</link><dc:creator>jjcm</dc:creator><comments>https://news.ycombinator.com/item?id=50399631</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50399631</guid></item><item><title><![CDATA[New comment by jjcm in "Flawless UI-UX Skill 1.0.0"]]></title><description><![CDATA[
<p>In general with these things, my thoughts are that one-shot/deslop skills should get you to a starting point quicker where you can iterate. Trying to go from 0->100 is usually a losing battle, as all of them will have elements that just don't quite work.<p>I do think impeccable steers away from slop / ai indicators the most. It aggressively removes standard AI thinks like eyebrows (present in the diffui render) or circle indicators (present in the flawless build). That doesn't mean the output is the best starting point however. It's a unique landing page, but not necessarily the best to convey or sell a waterproof folding phone. It comes at a cost however - impeccable leverages a <i>lot</i> of token spend to aggressively police itself (about $3.56, which is more than the diffui costs even with the image gen it did). Flawless was the cheapest of the three comparisons, but I don't see a significant difference between it and vanilla opus 5.5.</p>
]]></description><pubDate>Fri, 06 Nov 2026 20:10:38 +0000</pubDate><link>https://news.ycombinator.com/item?id=50397728</link><dc:creator>jjcm</dc:creator><comments>https://news.ycombinator.com/item?id=50397728</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50397728</guid></item><item><title><![CDATA[New comment by jjcm in "Flawless UI-UX Skill 1.0.0"]]></title><description><![CDATA[
<p>Examples comparing two similar de-slop skills. Flawless vs Impeccable vs Diffui (my own) for the same prompt[1]. Note that I did not prompt for mobile variants of these, so no clue how they'll show up on mobile.<p>Flawless (4.4M tokens): <a href="https://html.non.io/skill-compare-flawless/" rel="nofollow">https://html.non.io/skill-compare-flawless/</a><p>Impeccable (6.5M tokens): <a href="https://html.non.io/skill-compare-impeccable/" rel="nofollow">https://html.non.io/skill-compare-impeccable/</a><p>Diffui (2.4M tokens + $0.51 in api spend): <a href="https://html.non.io/skill-compare-diffui/" rel="nofollow">https://html.non.io/skill-compare-diffui/</a><p>[1] Full prompt:<p><pre><code>    Create three example webpage builds for the following prompt:

    "A landing page for a new foldable phone. Colorful dark green and gold theme. Should include a hero section, a build quality section emphasizing the waterproof capabilities, and a pricing section."

    Have three separate opus 5.5 cloud agents do this, each running in isolation.

    For the first use https://github.com/smokeeaasd/flawless-ui-ux

    For the second use https://impeccable.style

    For the third use https://diffui.ai</code></pre></p>
]]></description><pubDate>Fri, 06 Nov 2026 20:03:14 +0000</pubDate><link>https://news.ycombinator.com/item?id=50397632</link><dc:creator>jjcm</dc:creator><comments>https://news.ycombinator.com/item?id=50397632</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50397632</guid></item><item><title><![CDATA[New comment by jjcm in "Show HN: A decent text-to-image model trained for ~$20K (Apache 2.0 Weights)"]]></title><description><![CDATA[
<p>Very nice all. I've been playing around having poor results of training a pixel space image model of my own, though in a space very opposite from what you're doing (UI design, very heavy on text).<p>Any chance you'll release your training recipe as well? I know the model code is public, but I'd love to dive into the full process.</p>
]]></description><pubDate>Fri, 06 Nov 2026 04:07:45 +0000</pubDate><link>https://news.ycombinator.com/item?id=50388054</link><dc:creator>jjcm</dc:creator><comments>https://news.ycombinator.com/item?id=50388054</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50388054</guid></item></channel></rss>