<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: kgeist</title><link>https://news.ycombinator.com/user?id=kgeist</link><description>Hacker News RSS</description><docs>https://hnrss.org/</docs><generator>hnrss v2.1.1</generator><lastBuildDate>Sat, 21 Nov 2026 02:08:30 +0000</lastBuildDate><atom:link href="https://hnrss.org/user?id=kgeist" rel="self" type="application/rss+xml"></atom:link><item><title><![CDATA[New comment by kgeist in "What Comes After AGI?"]]></title><description><![CDATA[
<p>In 2024, most models were pretty bad at long horizon agentic tasks. I'd say it's mid/late 2025</p>
]]></description><pubDate>Fri, 20 Nov 2026 12:19:21 +0000</pubDate><link>https://news.ycombinator.com/item?id=50573701</link><dc:creator>kgeist</dc:creator><comments>https://news.ycombinator.com/item?id=50573701</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50573701</guid></item><item><title><![CDATA[New comment by kgeist in "Harvard researchers find AI coding agents shifted the bottleneck to code review"]]></title><description><![CDATA[
<p>Agree, many people think code reviews are there to catch "mistakes" (I used to think so too) but I think in most cases, they don't work that way. Bugs are still shipped and, at least in my experience, most reviewers often just quickly skim the changes without paying attention to the details. And those code reviews only give false confidence that a change is OK and must be error-free and good to go just because it was given LGTM by a senior dev (who barely looked at it), with all the consequences of such false confidence. It feels like in 95% of cases review is not even necessary and is just a corporate ritual at this point. It makes sense when you have a fresh contributor and you don't have confidence in them yet (i.e. code review as an onboarding process) or when a certain feature is non-trivial and you aren't sure everyone will be on the same page about the architecture (to prevent future conflicts/disagreements/rewrites of sorts), but other than that, proper QA processes are more important and more effective than CR at catching mistakes (tests, canaries, manual testing too etc.). Our company used to have a strong CR culture but now for a few years, there've been talks that CR is a bottleneck which doesn't really solve the problems it's supposed to solve.<p>My current pipeline with AI-generated code (and my own code too) is that I run AI reviewers several times on a single PR (the way LLMs work, due to attention optimizations, quantization issues, token sampling issues, harness issues -- AI agents often overlook some things on the first attempt), then I assemble the final report and we quickly decide together what needs to be addressed (critical stuff) and what's okay as is. I quickly skim the change set myself too, to see what the shape of the tests is, and what code areas were touched, just to be informed, and that's about it. I don't feel exhausted personally, because it pays off when there's a working feature at the end of the day</p>
]]></description><pubDate>Wed, 18 Nov 2026 04:29:16 +0000</pubDate><link>https://news.ycombinator.com/item?id=50541889</link><dc:creator>kgeist</dc:creator><comments>https://news.ycombinator.com/item?id=50541889</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50541889</guid></item><item><title><![CDATA[New comment by kgeist in "FBI says Artex AI open-source hacking tool used to breach networks"]]></title><description><![CDATA[
<p>Just a locally hosted GLM-5.3 in OpenCode.<p>Say, today I found a service had image uploads. I told it, go check that stuff. One of its hypotheses was a decompression bomb. 10 minutes later the service gives 503's and is stuck in CrashLoopBackoff. It has all the knowledge and tools to craft a bad PNG.<p>Or yesterday there was this article about Lovable users not securing their Supabase DB's properly. I told it to find a couple Lovable sites and probe them for exposed tables (no RLS). At first it refused to do it (the guardrails activated) but then it agreed to list only 1 row from each table with PII redacted as a compromise for my PoC. I then just opened a new session and asked it to remove PII redaction and return all rows. It complied becase it didn't pay attention to what the script was about. One of the Lovable sites exposed user passwords in plaintext.<p>Basically if my goal is "hack something successfully today" (by success I mean DoS, data breach or RCE), the success rate so far has been 100% (at least one site is compromised). It's scary how easy it is nowadays</p>
]]></description><pubDate>Mon, 16 Nov 2026 21:47:19 +0000</pubDate><link>https://news.ycombinator.com/item?id=50523783</link><dc:creator>kgeist</dc:creator><comments>https://news.ycombinator.com/item?id=50523783</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50523783</guid></item><item><title><![CDATA[New comment by kgeist in "FBI says Artex AI open-source hacking tool used to breach networks"]]></title><description><![CDATA[
<p>Not sure if you need this Artex tool at all? I was able to find 15 vulnerabilities including an RCE in 2 casual nights by just saying "go hack it" to a pretty good model with weak guardrails. No tooling needed except for a Linux environment</p>
]]></description><pubDate>Mon, 16 Nov 2026 21:20:02 +0000</pubDate><link>https://news.ycombinator.com/item?id=50523454</link><dc:creator>kgeist</dc:creator><comments>https://news.ycombinator.com/item?id=50523454</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50523454</guid></item><item><title><![CDATA[New comment by kgeist in "Serving GLM-5.3 at 607.8 tok/s across two Nvidia DGX stations with vLLM"]]></title><description><![CDATA[
<p>The setup in the article reports 600 tok/sec aggregate decode for GLM 5.3. I get the same throughput on 4x RTX 6000 Pro's connected vi PCIe for GLM 5.3 Flash, which is just 4 points below GLM 5.3 on Artificial Analysis. I'm just saying the setup in the article looks overpriced for what it can actually do</p>
]]></description><pubDate>Sat, 14 Nov 2026 19:42:27 +0000</pubDate><link>https://news.ycombinator.com/item?id=50500091</link><dc:creator>kgeist</dc:creator><comments>https://news.ycombinator.com/item?id=50500091</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50500091</guid></item><item><title><![CDATA[New comment by kgeist in "Serving GLM-5.3 at 607.8 tok/s across two Nvidia DGX stations with vLLM"]]></title><description><![CDATA[
<p>Dunno, maybe I'm good at prompting it, but I haven't had many issues with GLM-5.3 Flash. After switching from the Claude models, I barely noticed any difference for most tasks. In fact, I have to recheck the settings from time to time to make sure I'm not running anything Anthropic when I see it working with sensitive data :)<p>The only annoying issue for me is that GLM-5.3 Flash is often eager to go into action and start modifying code when I've just asked a question and it misunderstood it as a call to action.</p>
]]></description><pubDate>Sat, 14 Nov 2026 18:21:19 +0000</pubDate><link>https://news.ycombinator.com/item?id=50499382</link><dc:creator>kgeist</dc:creator><comments>https://news.ycombinator.com/item?id=50499382</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50499382</guid></item><item><title><![CDATA[New comment by kgeist in "Serving GLM-5.3 at 607.8 tok/s across two Nvidia DGX stations with vLLM"]]></title><description><![CDATA[
<p>People say GLM 5.3 is not much smarter than GLM 5.3 Flash (just +4 points on Artificial Analysis), and GLM 5.3 Flash fits on 4x RTX 6000 Pro for 4 times less money, with the same 600 tok/sec aggregate decode speed on vLLM (6 mln tokens KV cache). And I tell you, 600 tok/sec is not much for an organization, it's barely usable for 8 people. That's why we plan to buy more RTX 6000 Pros. And this setup requires $400,000 to run GLM 5.3 for 16 people, it's absolutely uneconomical.</p>
]]></description><pubDate>Sat, 14 Nov 2026 17:46:01 +0000</pubDate><link>https://news.ycombinator.com/item?id=50499050</link><dc:creator>kgeist</dc:creator><comments>https://news.ycombinator.com/item?id=50499050</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50499050</guid></item><item><title><![CDATA[New comment by kgeist in "Jensen Huang resigns"]]></title><description><![CDATA[
<p>>the real risk is the default world tech stack shifts to open models on Chinese chips.<p>As someone who isn't based in the US/EU and whose job is to deploy LLMs on premises, I can't wait for the moment when China starts selling its GPUs globally en masse with a proper CUDA equivalent. It's going to be cheaper in the long term and without all the export control drama.</p>
]]></description><pubDate>Thu, 12 Nov 2026 21:40:15 +0000</pubDate><link>https://news.ycombinator.com/item?id=50474690</link><dc:creator>kgeist</dc:creator><comments>https://news.ycombinator.com/item?id=50474690</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50474690</guid></item><item><title><![CDATA[New comment by kgeist in "Singapore Says U.S. Not Helpful on Investigation into Suspected Spy's Company"]]></title><description><![CDATA[
<p>>they can't match Nvidia + American tech stack for the foreseeable future.<p>>China can "just catch up", which begs the question why they don't have their own Nvidia already<p>What about Huawei and their Ascend chips?<p>DeepSeek and z.ai now run their models on Ascends, and inference is basically a solved problem (especially with the architectural/ML optimizations, like 90% of them come from China right now).<p>From what I understand, they lack compute for training - their own chips can do it too but they're slower and you need ~4 times more GPUs. Plus CUDA is a more mature stack than whatever Huawei provides (e.g. DeepSeek once failed to train on Ascends because they struggled with the stack). As long as China can buy Nvidia chips, it's cheaper and easier to just use smuggled chips but it doesn't mean they desperately need them.<p>>many of the automakers are operating on state subsidies<p>As opposed to many "successful" US companies living off investments without being actually profitable for years (or ever)?</p>
]]></description><pubDate>Thu, 12 Nov 2026 20:58:35 +0000</pubDate><link>https://news.ycombinator.com/item?id=50474181</link><dc:creator>kgeist</dc:creator><comments>https://news.ycombinator.com/item?id=50474181</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50474181</guid></item><item><title><![CDATA[New comment by kgeist in "Critical RCE Vulnerability in Apache HTTP Server"]]></title><description><![CDATA[
<p>But it's not a given that memory corruption is exploitable as an RCE specifically. I thought to call something an RCE you need a working exploit?</p>
]]></description><pubDate>Thu, 12 Nov 2026 03:19:52 +0000</pubDate><link>https://news.ycombinator.com/item?id=50462630</link><dc:creator>kgeist</dc:creator><comments>https://news.ycombinator.com/item?id=50462630</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50462630</guid></item><item><title><![CDATA[New comment by kgeist in "Critical RCE Vulnerability in Apache HTTP Server"]]></title><description><![CDATA[
<p>>Due to improper synchronization when executing concurrent threads sharing resources, an attacker can trigger heap corruption.<p>>As of the time of this blog’s publication, no exploit is available and there are no reports of exploitation in the wild.<p>Since when does "memory corruption" = "RCE"? Last week I found and reproduced a memory corruption (a use-after-free) in PHP's runtime after running an on-premise agent for just 30 minutes. There's nothing about it on the internet. Should I write a blog post about a scary new vulnerability? :)</p>
]]></description><pubDate>Thu, 12 Nov 2026 03:08:52 +0000</pubDate><link>https://news.ycombinator.com/item?id=50462559</link><dc:creator>kgeist</dc:creator><comments>https://news.ycombinator.com/item?id=50462559</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50462559</guid></item><item><title><![CDATA[New comment by kgeist in "Tracking Anthropic-discovered vulnerabilities"]]></title><description><![CDATA[
<p>Last week I deployed GLM-5.3-Flash on our on-premise servers, ran it against all repos, and in 6 hours it found 13 vulnerabilities across our products/sites, including 1 full account takeover and an RCE on our site which involved a pretty sophisticated chain of __destruct gadgets (all vulns verified to work by a human).<p>As an experiment, I also pointed it at the PHP runtime's source code, and it was able to construct PHP code that triggered a use-after-free (or a segfault, if you prefer DoS). I failed to find the vulnerability in the list of CVEs, so maybe it's a 0day, I don't know.<p>Sure, it's not a frontier model, but it can find many issues that were overlooked by all the commercial security scanners we run regularly as well. The nice thing about this setup is that we can run it 24/7 probing our code/infra, constructing exploits, etc. without having to care about token usage/bills. Plus, I think sending all that sensitive information about your internal infrastructure and the vulns found to Anthropic is a vulnerability in itself.</p>
]]></description><pubDate>Wed, 11 Nov 2026 07:59:45 +0000</pubDate><link>https://news.ycombinator.com/item?id=50450831</link><dc:creator>kgeist</dc:creator><comments>https://news.ycombinator.com/item?id=50450831</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50450831</guid></item><item><title><![CDATA[New comment by kgeist in "Llama-modchip: 284B DeepSeek on 64GB RAM, no GPU"]]></title><description><![CDATA[
<p>It takes 116 seconds to process 43 input tokens and produce an answer to a very simple request (translation). That's pretty unusable, to be honest. Is overthinking a trait of DeepSeek, or it's a low quantization issue?<p>For comparison, the whole thing takes 1.8 sec on my RTX 6000 rig with GLM5.3-Flash, and 12.5 seconds on my RTX 5090 rig with Qwen 3.6 27b (it overthinks too, unlike GLM).<p>I can't imagine waiting 2 minutes for each small turn. It's like... 1 coding task per night? Sounds like, as an engineer, I can accomplish more in that time if I did everything myself manually without any AI.</p>
]]></description><pubDate>Sun, 08 Nov 2026 21:41:17 +0000</pubDate><link>https://news.ycombinator.com/item?id=50417605</link><dc:creator>kgeist</dc:creator><comments>https://news.ycombinator.com/item?id=50417605</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50417605</guid></item><item><title><![CDATA[New comment by kgeist in "Preparing for the Flipper One Kickstarter and opening pre-orders"]]></title><description><![CDATA[
<p>The original developers are from Russia. They now seem to have an office in London</p>
]]></description><pubDate>Tue, 03 Nov 2026 20:20:56 +0000</pubDate><link>https://news.ycombinator.com/item?id=50357633</link><dc:creator>kgeist</dc:creator><comments>https://news.ycombinator.com/item?id=50357633</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50357633</guid></item><item><title><![CDATA[New comment by kgeist in "Revolut moves to self-hosted GLM"]]></title><description><![CDATA[
<p>I'm currently evaluating GLM and oh boy, lots of correctness issues on large contexts if you run it on vLLM (bugs in the engine itself). I wonder what their config is.</p>
]]></description><pubDate>Tue, 03 Nov 2026 15:01:21 +0000</pubDate><link>https://news.ycombinator.com/item?id=50352859</link><dc:creator>kgeist</dc:creator><comments>https://news.ycombinator.com/item?id=50352859</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50352859</guid></item><item><title><![CDATA[New comment by kgeist in "When Modernization Breaks Working Code: A Study of LLMs on Legacy PHP"]]></title><description><![CDATA[
<p>The average age of the explored models is ~1.5 years. They discuss PHP 8.3, which was released in 2023 and has not been actively supported since 2025. Agentic coding isn't used.<p>I'm not sure what the point is of publishing results on the mid-2025 state of the art in October 2026.</p>
]]></description><pubDate>Mon, 02 Nov 2026 23:42:32 +0000</pubDate><link>https://news.ycombinator.com/item?id=50345843</link><dc:creator>kgeist</dc:creator><comments>https://news.ycombinator.com/item?id=50345843</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50345843</guid></item><item><title><![CDATA[New comment by kgeist in "OpenAI's agents talked to each other through the CPU Cache"]]></title><description><![CDATA[
<p>>This misses the point: AIs will get better and better, and containment will be harder and harder.<p>Containment is hard when you give an AI agent access to a whole VM with a huge attack surface.<p>And it's pretty trivial when you give it only a small whitelist of manually curated/reviewed/tested primitives.</p>
]]></description><pubDate>Thu, 29 Oct 2026 10:25:07 +0000</pubDate><link>https://news.ycombinator.com/item?id=50286869</link><dc:creator>kgeist</dc:creator><comments>https://news.ycombinator.com/item?id=50286869</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50286869</guid></item><item><title><![CDATA[New comment by kgeist in "There isn't an easy exploit for this PHP vulnerability"]]></title><description><![CDATA[
<p>I remember our infosec person instructing us to verify we don't use log4j in our system written in PHP. Well, we didn't :)</p>
]]></description><pubDate>Thu, 22 Oct 2026 02:42:44 +0000</pubDate><link>https://news.ycombinator.com/item?id=50190526</link><dc:creator>kgeist</dc:creator><comments>https://news.ycombinator.com/item?id=50190526</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50190526</guid></item><item><title><![CDATA[New comment by kgeist in "There isn't an easy exploit for this PHP vulnerability"]]></title><description><![CDATA[
<p>I ran 2 PoCs from GitHub today: one indeed expected Local File Inclusion (LFI) to already be present (so it's just LFI with extra steps), and the other one didn't work at all out of the box without some modifications (although I used the same Docker images and all). The second PoC did trigger a real stale write bug in the end, but it didn't demonstrate actual RCE; it just triggered a stale write. And then I wondered, is this the first time ever that the PHP runtime has had a memory corruption bug? I doubt it (I remember a few years ago we had random crashes in production under high load due to an opcache bug). And if it's not the first memory corruption bug in the PHP runtime, why is it being proclaimed the worst vulnerability in years?</p>
]]></description><pubDate>Thu, 22 Oct 2026 00:29:21 +0000</pubDate><link>https://news.ycombinator.com/item?id=50189528</link><dc:creator>kgeist</dc:creator><comments>https://news.ycombinator.com/item?id=50189528</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50189528</guid></item><item><title><![CDATA[New comment by kgeist in "Show HN: Llmspeed – feel how fast a local LLM runs before you buy the hardware"]]></title><description><![CDATA[
<p>It says 46 tok/sec for Qwen3.6-27b 4 bit on RTX5090, but I get 60-70 tok/sec on a small context like that (5 bit weights with 8 bit KV cache), although it does slow down to 40 tok/sec on large contexts. If I add ngram speculation, then it can increase to 100+ tok/sec on code.<p>There are many knobs to tweak to make it faster, which the site ignores.</p>
]]></description><pubDate>Sat, 17 Oct 2026 20:18:36 +0000</pubDate><link>https://news.ycombinator.com/item?id=50134730</link><dc:creator>kgeist</dc:creator><comments>https://news.ycombinator.com/item?id=50134730</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50134730</guid></item></channel></rss>