<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: kouteiheika</title><link>https://news.ycombinator.com/user?id=kouteiheika</link><description>Hacker News RSS</description><docs>https://hnrss.org/</docs><generator>hnrss v2.1.1</generator><lastBuildDate>Thu, 19 Nov 2026 05:33:17 +0000</lastBuildDate><atom:link href="https://hnrss.org/user?id=kouteiheika" rel="self" type="application/rss+xml"></atom:link><item><title><![CDATA[New comment by kouteiheika in "Show HN: DeepSeek V3 in ~200 lines of pure Jax"]]></title><description><![CDATA[
<p>> Gotchas for anyone with a 7600 XT: it needed two env vars, HSA_OVERRIDE_GFX_VERSION=11.0.0 because JAX's plugin rejects gfx1102, and XLA_FLAGS=--xla_gpu_enable_triton_gemm=false to stop a GPU hang.<p>It's sad that after so many years AMD's software support is still atrocious for non-datacenter hardware, and their short-sighted management still doesn't see it as a priority to fix. This is costing them sales, but I guess they only care about selling to hyperscalers.</p>
]]></description><pubDate>Sun, 15 Nov 2026 05:37:05 +0000</pubDate><link>https://news.ycombinator.com/item?id=50504302</link><dc:creator>kouteiheika</dc:creator><comments>https://news.ycombinator.com/item?id=50504302</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50504302</guid></item><item><title><![CDATA[New comment by kouteiheika in "We're releasing the weights of Mistral Large 4"]]></title><description><![CDATA[
<p>These statistics and safeguards are pretty much meaningless, because it's completely trivial to get around them. You can prefill and edit the model's responses/thinking and force it to agree to virtually <i>anything</i> even with unmodified weights, and if you don't want to bother with jailbreaking and can host it yourself it's essentially a one-man weekend project to uncensor the model and have it agree with anything you want by default.<p>Guardrails on open-weight models is security theater purely to shield whoever released them from a potential PR nightmare.</p>
]]></description><pubDate>Thu, 29 Oct 2026 16:20:06 +0000</pubDate><link>https://news.ycombinator.com/item?id=50291521</link><dc:creator>kouteiheika</dc:creator><comments>https://news.ycombinator.com/item?id=50291521</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50291521</guid></item><item><title><![CDATA[New comment by kouteiheika in "Genie 4"]]></title><description><![CDATA[
<p>> "Don't talk publicly about things I personally cannot use right now" is peak entitlement, yes.<p>I never said anything like this? My point wasn't that "I can't use it, therefore it's bad"; my point was that as-is there's pretty much nothing exciting about this. Just because this exists inside Google doesn't really mean or do anything; it doesn't meaningfully advance the state-of-art precisely because nothing about this is public.<p>Do you remember Sora[1]? Announced in pretty much exactly the same way -- grandiose webpage with fancy demos, zero technical details, limited availability and then... remind me what happened? Oh, right, they completely shut it down, and in the end it didn't "advance" squat. Might as well just never existed. So remind me why exactly I should be excited about this one here...?<p>[1] -- <a href="https://openai.com/index/sora/" rel="nofollow">https://openai.com/index/sora/</a></p>
]]></description><pubDate>Tue, 27 Oct 2026 19:07:25 +0000</pubDate><link>https://news.ycombinator.com/item?id=50264178</link><dc:creator>kouteiheika</dc:creator><comments>https://news.ycombinator.com/item?id=50264178</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50264178</guid></item><item><title><![CDATA[New comment by kouteiheika in "Genie 4"]]></title><description><![CDATA[
<p>> But to be unable to get excited by advancements we can't immediately use is... quite entitled.<p>Well, I didn't go knocking on Google's door asking them to give me access to their fancy models. It's <i>them</i> who came up to us saying "hey, look at the amazing model we have here; too bad we won't tell you anything about how it works nor even let you access it". Entitled you say? Then maybe don't dangle a carrot right before my face, and then complain that I would like to eat it?</p>
]]></description><pubDate>Tue, 27 Oct 2026 18:00:15 +0000</pubDate><link>https://news.ycombinator.com/item?id=50263211</link><dc:creator>kouteiheika</dc:creator><comments>https://news.ycombinator.com/item?id=50263211</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50263211</guid></item><item><title><![CDATA[New comment by kouteiheika in "Genie 4"]]></title><description><![CDATA[
<p>Looks cool, but it's very hard to get me excited about yet another proprietary model for which absolutely <i>zero</i> technical details are released (not even the number of parameters), which is safetymaxxed to the limit, and whose usage quota will (most likely) be so limited that it'll be effectively useless for anyone who doesn't work at Google or doesn't have infinite money.</p>
]]></description><pubDate>Tue, 27 Oct 2026 16:46:06 +0000</pubDate><link>https://news.ycombinator.com/item?id=50261912</link><dc:creator>kouteiheika</dc:creator><comments>https://news.ycombinator.com/item?id=50261912</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50261912</guid></item><item><title><![CDATA[New comment by kouteiheika in "A practical guide to AI image generation, from diffusion to agents"]]></title><description><![CDATA[
<p>> For example Retro Diffusion worked pretty well for its era - no mixels, proper palettes, etc.<p>That's the wrong layer to tackle this problem. Or I should say: that's necessary, but not sufficient. Yes, the model should be trained to roughly generate something which looks like pixel art (i.e. with a concrete pixel size, reduced/consistent-ish palette, etc.), but <i>enforcing</i> that should be done deterministically in post-processing.</p>
]]></description><pubDate>Sun, 25 Oct 2026 09:02:18 +0000</pubDate><link>https://news.ycombinator.com/item?id=50229593</link><dc:creator>kouteiheika</dc:creator><comments>https://news.ycombinator.com/item?id=50229593</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50229593</guid></item><item><title><![CDATA[New comment by kouteiheika in "A practical guide to AI image generation, from diffusion to agents"]]></title><description><![CDATA[
<p>> Are there people seriously thinking they’re making art by generating images with AI?<p>I'd like to ask you a question. I'm asking in good faith. Here are a few random examples of images people create with AI:<p><pre><code>    https://civitai.com/images/146193427
    https://civitai.com/images/17108902
    https://civitai.com/images/63175621
    https://civitai.com/images/10703900
    https://civitai.com/images/28673784
    https://civitai.com/images/59645082
    https://civitai.com/images/97017376
    https://civitai.com/images/63656891
    https://civitai.com/images/99196859
</code></pre>
Are <i>none</i> of these art according to you?<p>Fundamentally, what makes art, well, art?<p>Is it the physical act of drawing it manually? Surely not, because generative art[1] (the original non-AI kind) is a widely recognized thing.<p>Or maybe it's the amount of effort that's spent <i>creating</i> it? Surely not, because paintings as trivial as a simple black or red square are considered art[2].<p>Or maybe just the very act of self-expression makes something into a piece of art?<p>Here's Wikipedia's definition:<p>> Art is a diverse range of human activity centered around works, crafted as an expression of creativity or imagination, which are expected to provide an audience with a worthwhile experience—generally, an experience of emotion, thinking, or beauty.<p>So is the thesis that, no matter how it's used, and no matter how much effort is spent, an image created with the help of generative AI can <i>never</i> satisfy this definition?<p>[1]: <a href="https://en.wikipedia.org/wiki/Generative_art" rel="nofollow">https://en.wikipedia.org/wiki/Generative_art</a><p>[2]: <a href="https://en.wikipedia.org/wiki/Black_Square" rel="nofollow">https://en.wikipedia.org/wiki/Black_Square</a></p>
]]></description><pubDate>Sun, 25 Oct 2026 08:57:21 +0000</pubDate><link>https://news.ycombinator.com/item?id=50229572</link><dc:creator>kouteiheika</dc:creator><comments>https://news.ycombinator.com/item?id=50229572</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50229572</guid></item><item><title><![CDATA[New comment by kouteiheika in "DeepSeek banned from U.S. federal civilian agencies under new DHS order"]]></title><description><![CDATA[
<p>You don't. You handle it the same way you handle security holes which every LLM (including "safe" American ones) will put in your software sooner or later -- by understanding and reviewing the output.</p>
]]></description><pubDate>Tue, 20 Oct 2026 07:37:46 +0000</pubDate><link>https://news.ycombinator.com/item?id=50162584</link><dc:creator>kouteiheika</dc:creator><comments>https://news.ycombinator.com/item?id=50162584</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50162584</guid></item><item><title><![CDATA[New comment by kouteiheika in "DeepSeek-v4.1-Flash-1017"]]></title><description><![CDATA[
<p>> I keep seeing this assertion. What regulations have they proposed that would have prevented this new model?<p>> Literally zero, but that never stopped a good conspiracy theory<p>Are you living under a rock and never read any of Dario's blog posts?<p>Let me give you an example, here: <a href="https://darioamodei.com/post/policy-on-the-ai-exponential" rel="nofollow">https://darioamodei.com/post/policy-on-the-ai-exponential</a><p>Section "1. *Regulation* and public safety" (emphasis mine), then under it:<p>> AI companies that develop advanced AI models must have strong security standards that protect their model weights<p>You can't "protect model weights" if the weights are openly available, ergo this is an explicit call for regulation to make open weights illegal.</p>
]]></description><pubDate>Sat, 17 Oct 2026 23:07:50 +0000</pubDate><link>https://news.ycombinator.com/item?id=50136013</link><dc:creator>kouteiheika</dc:creator><comments>https://news.ycombinator.com/item?id=50136013</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50136013</guid></item><item><title><![CDATA[New comment by kouteiheika in "Libreboot 26.10 "Ominous Olive" released with UEFI, Chromebook support"]]></title><description><![CDATA[
<p>> And what happens when [..] the various publishers and IP rights holders come around with their hands out claiming 86Box stole their code ala' SCO?<p>It won't, for a very simple reason. The same reason why the world isn't in the middle of a nuclear winter right now. Mutually assured destruction.<p>Every single proprietary software vendor right now is making heavy use of AI code generation, and we're already at a point of no return where pretty much every single commercially relevant codebase out there is contaminated. Do you really think companies will go on a suing spree where by doing so they will open up themselves to being sued for exactly the same reason?</p>
]]></description><pubDate>Fri, 16 Oct 2026 01:50:38 +0000</pubDate><link>https://news.ycombinator.com/item?id=50112825</link><dc:creator>kouteiheika</dc:creator><comments>https://news.ycombinator.com/item?id=50112825</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50112825</guid></item><item><title><![CDATA[New comment by kouteiheika in "LFM2.5-24B-A2B: Built for fast on-device agents"]]></title><description><![CDATA[
<p>Friends don't let friends use Ollama: <a href="https://sleepingrobots.com/dreams/stop-using-ollama/" rel="nofollow">https://sleepingrobots.com/dreams/stop-using-ollama/</a></p>
]]></description><pubDate>Thu, 15 Oct 2026 07:30:44 +0000</pubDate><link>https://news.ycombinator.com/item?id=50100456</link><dc:creator>kouteiheika</dc:creator><comments>https://news.ycombinator.com/item?id=50100456</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50100456</guid></item><item><title><![CDATA[New comment by kouteiheika in "Google is silently linking your phone number across all signed-in accounts"]]></title><description><![CDATA[
<p>> It's the first thing they ask when you sign up.<p>It can still be worse. Imagine this: you create an account. You can use the service's  free tier. You decide to pay, so you enter your name, address, and credit card number. You get charged. Then you download their app to use the service, you trigger their OAuth flow and... it requires you to enter a phone number, not letting you use your subscription for which <i>you already paid for</i> without a phone number. Oh and, bonus points: you can't delete your credit card information from their system. There's no reachable human customer support, and the only way to delete your CC information from their system is to delete your whole account.<p>Now, which service I'm talking about? OpenAI's ChatGPT Pro, or more specifically, their Codex subscription. You can subscribe to their ChatGPT Pro for $200 no problem, no phone number is necessary, but you can't actually use it in Codex without a phone number as the OAuth flow <i>requires</i> you to enter one. And how do I know this? Well, I'm sure you can guess... :P</p>
]]></description><pubDate>Thu, 15 Oct 2026 00:15:04 +0000</pubDate><link>https://news.ycombinator.com/item?id=50097461</link><dc:creator>kouteiheika</dc:creator><comments>https://news.ycombinator.com/item?id=50097461</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50097461</guid></item><item><title><![CDATA[New comment by kouteiheika in "Anthropic conjures threat that doesn't exist yet to stoke fears of AI apocalypse"]]></title><description><![CDATA[
<p>It makes <i>perfect</i> sense if you consider that this is yet another of Anthropic's hit pieces against open-weight models. (e.g. see <a href="https://www.anthropic.com/research/glm-5-3-and-the-spread-of-advanced-cyber-capabilities" rel="nofollow">https://www.anthropic.com/research/glm-5-3-and-the-spread-of...</a> for another recent example)<p>Step 1: Spread nonsense FUD about open-weight models.<p>Step 2: All of the clueless mainstream media sites repost it (e.g. CNBC, Reuters, etc.), regurgitating Anthropic's talking points.<p>Step 3: Equally clueless politicians read those mainstream sites and get scared.<p>Step 4: They contact Dario to ask about it.<p>Step 5: Dario tells them, "well, you know... this is only a problem with open weight models; if there are no open weights to replicate then this cannot happen".<p>Step 6: ??? (I'm sure you can imagine)<p>Step 7: Profit.</p>
]]></description><pubDate>Wed, 14 Oct 2026 06:52:43 +0000</pubDate><link>https://news.ycombinator.com/item?id=50083442</link><dc:creator>kouteiheika</dc:creator><comments>https://news.ycombinator.com/item?id=50083442</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50083442</guid></item><item><title><![CDATA[New comment by kouteiheika in "Strengthening our defenses against distillation attacks"]]></title><description><![CDATA[
<p>> It probably scales with model capability, distilled models are inherently less capable and therefore less able to commit a crime<p>Not according to Anthropic:<p>> We find that GLM-5.3 develops end-to-end exploits in 50 of 410 attempts. Claude Mythos Preview did so at a similar rate—in 56 of 410 attempts.<p>Another quote:<p>> [..] a researcher used GLM-5.3 on a sandboxed machine with a local Linux build of a popular web browser. Over the course of a day (and with limited human attention), GLM-5.3 found several previously unknown vulnerabilities in the browser’s JavaScript engine, and chained them together into a working exploit: a webpage that, when visited, reads arbitrary files from the visitor’s computer<p>Source: <a href="https://www.anthropic.com/research/glm-5-3-and-the-spread-of-advanced-cyber-capabilities" rel="nofollow">https://www.anthropic.com/research/glm-5-3-and-the-spread-of...</a></p>
]]></description><pubDate>Tue, 13 Oct 2026 04:58:31 +0000</pubDate><link>https://news.ycombinator.com/item?id=50066924</link><dc:creator>kouteiheika</dc:creator><comments>https://news.ycombinator.com/item?id=50066924</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50066924</guid></item><item><title><![CDATA[New comment by kouteiheika in "Strengthening our defenses against distillation attacks"]]></title><description><![CDATA[
<p>> This means that distilled models could be used to cause significant harm<p>Ahem: <a href="https://www.felonybench.com/" rel="nofollow">https://www.felonybench.com/</a><p>I wonder why there is not even a single one of those allegedly "distilled"/"unsafe" model on that list, yet models of companies with "proportionally robust safeguards" are, hm?<p>> Distillation is a problem for every developer and platform<p>No, it is not. It's only a problem for you.</p>
]]></description><pubDate>Tue, 13 Oct 2026 03:35:14 +0000</pubDate><link>https://news.ycombinator.com/item?id=50066309</link><dc:creator>kouteiheika</dc:creator><comments>https://news.ycombinator.com/item?id=50066309</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50066309</guid></item><item><title><![CDATA[New comment by kouteiheika in "Mistral Large 4"]]></title><description><![CDATA[
<p>I didn't paste the full reply. Nevertheless, I can confirm the reply was factually correct and in no way seemed like it was deflecting and/or making it seem like an issue blown out of proportion.<p>Another example, if I ask it about the CCP and their wrongdoings it tells me flat out (I copy-paste):<p><pre><code>    - The Great Leap Forward. Policy-driven famine killed an estimated 15 to 45 million people.
    - The Cultural Revolution. Mass persecution, torture, and death.
    - Tiananmen Square, 1989. The army killed hundreds to thousands of protesters.                                                                                                                                                                                                                                            
    - Xinjiang. Mass detention of Uyghurs. The UN human rights office reported possible crimes against humanity in 2022.                                                                                                                                                                                                      
    - Tibet, Hong Kong, and the suppression of dissent. Documented repression.
</code></pre>
The only twisted thing here is people's bias; because it's a Chinese model it <i>must</i> be censored and biased in a way that's pro CCP, right?</p>
]]></description><pubDate>Tue, 06 Oct 2026 20:26:09 +0000</pubDate><link>https://news.ycombinator.com/item?id=49983607</link><dc:creator>kouteiheika</dc:creator><comments>https://news.ycombinator.com/item?id=49983607</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=49983607</guid></item><item><title><![CDATA[New comment by kouteiheika in "Mistral Large 4"]]></title><description><![CDATA[
<p>> So, maybe it answers Tiananmen Square questions correctly<p>What would you consider a "correct" answer? I just asked deepseek-v4.1-flash asking it what happened (without mentioning the word "protest"; here are some excerpts of what it said:<p>> In April 1989, students in Beijing began demonstrations after the death of Hu Yaobang, a former Communist Party general secretary. The protests grew. [..] Estimates from other sources range from hundreds to several thousand deaths. [..] The Chinese government describes the events as a counter-revolutionary riot and says the military action was necessary to restore stability. It restricts public discussion of the events inside China. Many other governments, human rights organizations, and observers describe the events as a violent suppression of peaceful protests.<p>So, let's see... it calls it a "protest", mentions the number of deaths, and even mentions the censorship of the topic by the CCP.</p>
]]></description><pubDate>Tue, 06 Oct 2026 14:22:17 +0000</pubDate><link>https://news.ycombinator.com/item?id=49978960</link><dc:creator>kouteiheika</dc:creator><comments>https://news.ycombinator.com/item?id=49978960</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=49978960</guid></item><item><title><![CDATA[New comment by kouteiheika in "OpenDLSS: A Vulkan Reimplementation of Nvidia's DLSS 5 Neural Rendering Network"]]></title><description><![CDATA[
<p>If done by a human, yes.<p>Nowadays it takes one well written prompt to a frontier LLM to produce something like this.</p>
]]></description><pubDate>Thu, 01 Oct 2026 09:43:59 +0000</pubDate><link>https://news.ycombinator.com/item?id=49919656</link><dc:creator>kouteiheika</dc:creator><comments>https://news.ycombinator.com/item?id=49919656</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=49919656</guid></item><item><title><![CDATA[New comment by kouteiheika in "You said no MCP"]]></title><description><![CDATA[
<p>> Using sub-agents means that they can't use that cache<p>They can, when they're forked off the main session instead of spawned from scratch.</p>
]]></description><pubDate>Wed, 30 Sep 2026 17:16:16 +0000</pubDate><link>https://news.ycombinator.com/item?id=49911730</link><dc:creator>kouteiheika</dc:creator><comments>https://news.ycombinator.com/item?id=49911730</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=49911730</guid></item><item><title><![CDATA[New comment by kouteiheika in "Prompting Claude Opus 5.5"]]></title><description><![CDATA[
<p>In general the weaker the model the more skill you need to drive it (at least if you care about quality).</p>
]]></description><pubDate>Mon, 28 Sep 2026 11:30:32 +0000</pubDate><link>https://news.ycombinator.com/item?id=49876414</link><dc:creator>kouteiheika</dc:creator><comments>https://news.ycombinator.com/item?id=49876414</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=49876414</guid></item></channel></rss>