<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: seanmcdirmid</title><link>https://news.ycombinator.com/user?id=seanmcdirmid</link><description>Hacker News RSS</description><docs>https://hnrss.org/</docs><generator>hnrss v2.1.1</generator><lastBuildDate>Mon, 17 Aug 2026 13:03:06 +0000</lastBuildDate><atom:link href="https://hnrss.org/user?id=seanmcdirmid" rel="self" type="application/rss+xml"></atom:link><item><title><![CDATA[New comment by seanmcdirmid in "Abdominal fat predicts heart disease risk better than BMI"]]></title><description><![CDATA[
<p>You can induce a calorie deficit by eating foods that fill you up but don’t have many calories. Also avoiding the sugar-spike carbs cause leaving you even hungrier shortly after works also.<p>A good dietician is worth a few consults if you don’t have the will power to just fast. And as a bonus you don’t lose muscle mass as well because they’ll focus on keeping protein up.</p>
]]></description><pubDate>Sun, 16 Aug 2026 05:24:44 +0000</pubDate><link>https://news.ycombinator.com/item?id=49317159</link><dc:creator>seanmcdirmid</dc:creator><comments>https://news.ycombinator.com/item?id=49317159</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=49317159</guid></item><item><title><![CDATA[New comment by seanmcdirmid in "Asus Bike Booster"]]></title><description><![CDATA[
<p>Only because there are so many targets to steal from. And by Asia you mean China, where E bikes are rarely very fancy and often the size of a moped (except for the ride shares, which are still bricks and GPS tracked). And judging by the noise level at the hotel I stayed at in Chengdu a couple of weeks ago, they definitely come with theft alarms.</p>
]]></description><pubDate>Sun, 16 Aug 2026 05:15:48 +0000</pubDate><link>https://news.ycombinator.com/item?id=49317121</link><dc:creator>seanmcdirmid</dc:creator><comments>https://news.ycombinator.com/item?id=49317121</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=49317121</guid></item><item><title><![CDATA[New comment by seanmcdirmid in "Magnitude 7.7 Earthquake – 68 km NNW of Ende, Indonesia"]]></title><description><![CDATA[
<p>Lombok will have lots of higher ground at least. You really just have to be careful on and near the beach.</p>
]]></description><pubDate>Sat, 15 Aug 2026 04:58:36 +0000</pubDate><link>https://news.ycombinator.com/item?id=49307771</link><dc:creator>seanmcdirmid</dc:creator><comments>https://news.ycombinator.com/item?id=49307771</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=49307771</guid></item><item><title><![CDATA[New comment by seanmcdirmid in "Qwen 3.8 27B"]]></title><description><![CDATA[
<p>True. Benchmarks have led me astray before, I'll have to actually benchmark it directly in Goose with my real use cases right now (which are mostly writing python code, so it probably still doesn't apply?). Still, I've never noticed a benefit to using a dense model rather than an MoE for coding with an agent tasks.</p>
]]></description><pubDate>Fri, 14 Aug 2026 21:29:23 +0000</pubDate><link>https://news.ycombinator.com/item?id=49304821</link><dc:creator>seanmcdirmid</dc:creator><comments>https://news.ycombinator.com/item?id=49304821</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=49304821</guid></item><item><title><![CDATA[New comment by seanmcdirmid in "Qwen 3.8 27B"]]></title><description><![CDATA[
<p>My use cases try to avoid accessing world knowledge in the model (I give it access to web search for some adhoc RAG), and ya, I'm just focused on coding so that's the only place I'm looking at right now.</p>
]]></description><pubDate>Fri, 14 Aug 2026 21:07:26 +0000</pubDate><link>https://news.ycombinator.com/item?id=49304609</link><dc:creator>seanmcdirmid</dc:creator><comments>https://news.ycombinator.com/item?id=49304609</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=49304609</guid></item><item><title><![CDATA[New comment by seanmcdirmid in "Qwen 3.8 27B"]]></title><description><![CDATA[
<p>I'm struggling to figure out what to use this for. From the intelligence benchmarks in OMLX. If only they would release another MoE model.<p><pre><code>  Intelligence Benchmark Comparison

  --- Detail ---

  Model: scottlowry--Qwen3.8-27B-oQ4e-mtp
  Benchmark         Accuracy   Correct   Total   Time(s)   Think
  --------------------------------------------------------------
  GSM8K                93.3%        28      30       282      No
  MATHQA               46.7%        14      30      26.3      No
  HUMANEVAL            96.7%        29      30     156.5      No
  MBPP                 83.3%        25      30      71.5      No
  LIVECODEBENCH        43.3%        13      30    1040.4      No

  Model: stamsam--Qwen3.6-35B-A3B-Claude-4.7-Opus-Reasoning-Distilled-MLX-oQ4-MTP
  Benchmark         Accuracy   Correct   Total   Time(s)   Think
  --------------------------------------------------------------
  GSM8K                96.7%        29      30      51.9      No
  MATHQA               60.0%        18      30       9.1      No
  HUMANEVAL            83.3%        25      30      82.9      No
  MBPP                 80.0%        24      30      29.6      No
  LIVECODEBENCH        36.7%        11      30     283.7      No</code></pre></p>
]]></description><pubDate>Fri, 14 Aug 2026 20:31:59 +0000</pubDate><link>https://news.ycombinator.com/item?id=49304196</link><dc:creator>seanmcdirmid</dc:creator><comments>https://news.ycombinator.com/item?id=49304196</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=49304196</guid></item><item><title><![CDATA[New comment by seanmcdirmid in "DeepSeek API Pricing Update"]]></title><description><![CDATA[
<p>DeepSeek still has the best prices on offer directly. I wouldn't go with a US provider unless latency or privacy was an issue.</p>
]]></description><pubDate>Fri, 14 Aug 2026 17:58:24 +0000</pubDate><link>https://news.ycombinator.com/item?id=49302337</link><dc:creator>seanmcdirmid</dc:creator><comments>https://news.ycombinator.com/item?id=49302337</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=49302337</guid></item><item><title><![CDATA[New comment by seanmcdirmid in "Qwen 3.8 27B"]]></title><description><![CDATA[
<p>not parent, but 4-bit quantization is generally consider a good trade off for speed/performance, so you might use it even when you aren't on consumer hardware, but definitely when you are on consumer hardware.</p>
]]></description><pubDate>Fri, 14 Aug 2026 17:51:46 +0000</pubDate><link>https://news.ycombinator.com/item?id=49302227</link><dc:creator>seanmcdirmid</dc:creator><comments>https://news.ycombinator.com/item?id=49302227</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=49302227</guid></item><item><title><![CDATA[New comment by seanmcdirmid in "Qwen3.8-27B"]]></title><description><![CDATA[
<p>27B is a dense model so it will be slower with an MoE (A3B), but should have better quality? I still haven’t found very good uses cases on my M3 Max for dense models. Even if you can find a MTP version, it doesn’t help much, especially if you compare against an MoE with MTP as well.</p>
]]></description><pubDate>Fri, 14 Aug 2026 16:41:57 +0000</pubDate><link>https://news.ycombinator.com/item?id=49301196</link><dc:creator>seanmcdirmid</dc:creator><comments>https://news.ycombinator.com/item?id=49301196</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=49301196</guid></item><item><title><![CDATA[New comment by seanmcdirmid in "Deutsche Bank becomes first foreign yuan clearing bank in Europe"]]></title><description><![CDATA[
<p>One advantage America has, and will remain having as an advantage as long as its broken: it borrows a lot of dollars. So if China wants to save a few billion dollars, it is super easy to do it by just bidding for US treasuries, rather than trying to convert it to yuan and somehow putting it in its own economy where it won't cause overheating.<p>America is essentially just a debtor of last resort, that is the superpower of the dollar.</p>
]]></description><pubDate>Fri, 14 Aug 2026 05:09:25 +0000</pubDate><link>https://news.ycombinator.com/item?id=49294937</link><dc:creator>seanmcdirmid</dc:creator><comments>https://news.ycombinator.com/item?id=49294937</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=49294937</guid></item><item><title><![CDATA[New comment by seanmcdirmid in "Muse Glimmer: 30B-parameter model optimized for always-on local agent workflows"]]></title><description><![CDATA[
<p>Alibaba probably. They are doing work on both already, that is if a quantum LLM is even feasible.</p>
]]></description><pubDate>Mon, 10 Aug 2026 22:46:31 +0000</pubDate><link>https://news.ycombinator.com/item?id=49250938</link><dc:creator>seanmcdirmid</dc:creator><comments>https://news.ycombinator.com/item?id=49250938</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=49250938</guid></item><item><title><![CDATA[New comment by seanmcdirmid in "Two wheels, a few tradeoffs, and gas prices"]]></title><description><![CDATA[
<p>You still have to make a call on whether it’s worth it or not. People who think price doesn’t matter don’t live very long anyways.</p>
]]></description><pubDate>Mon, 10 Aug 2026 19:49:58 +0000</pubDate><link>https://news.ycombinator.com/item?id=49248776</link><dc:creator>seanmcdirmid</dc:creator><comments>https://news.ycombinator.com/item?id=49248776</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=49248776</guid></item><item><title><![CDATA[New comment by seanmcdirmid in "Two wheels, a few tradeoffs, and gas prices"]]></title><description><![CDATA[
<p>Dying as a motorcyclist is different from dying. Life insurance either pays off or it doesn’t, if you hold it from 20 to 90, it usually doesn’t pay off.</p>
]]></description><pubDate>Mon, 10 Aug 2026 16:29:20 +0000</pubDate><link>https://news.ycombinator.com/item?id=49245943</link><dc:creator>seanmcdirmid</dc:creator><comments>https://news.ycombinator.com/item?id=49245943</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=49245943</guid></item><item><title><![CDATA[New comment by seanmcdirmid in "Muse Glimmer: 30B-parameter model optimized for always-on local agent workflows"]]></title><description><![CDATA[
<p>You can compare Qwen with thinking to Qwen with no thinking though. I find my results are better without thinking because of overthinking.</p>
]]></description><pubDate>Mon, 10 Aug 2026 16:27:08 +0000</pubDate><link>https://news.ycombinator.com/item?id=49245899</link><dc:creator>seanmcdirmid</dc:creator><comments>https://news.ycombinator.com/item?id=49245899</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=49245899</guid></item><item><title><![CDATA[New comment by seanmcdirmid in "Meta Muse Glimmer – Open weights 30B local coding model"]]></title><description><![CDATA[
<p>Not parent, but I use Goose for my non-handcrafted Qwen use cases, I’m also working on handcrafting as well. Goose was the only harness that didnt bloat context too much with system prompts (like openclaw) and I could get reasonable web search working with Qwen.</p>
]]></description><pubDate>Mon, 10 Aug 2026 16:25:49 +0000</pubDate><link>https://news.ycombinator.com/item?id=49245878</link><dc:creator>seanmcdirmid</dc:creator><comments>https://news.ycombinator.com/item?id=49245878</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=49245878</guid></item><item><title><![CDATA[New comment by seanmcdirmid in "Meta Muse Glimmer – open weights 30B local coding model"]]></title><description><![CDATA[
<p>Disable thinking? I think many harnesses disable thinking on Qwen anyways because it interferes with tool calling.</p>
]]></description><pubDate>Mon, 10 Aug 2026 16:23:57 +0000</pubDate><link>https://news.ycombinator.com/item?id=49245845</link><dc:creator>seanmcdirmid</dc:creator><comments>https://news.ycombinator.com/item?id=49245845</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=49245845</guid></item><item><title><![CDATA[New comment by seanmcdirmid in "Two wheels, a few tradeoffs, and gas prices"]]></title><description><![CDATA[
<p>Existentials are always true trivially.</p>
]]></description><pubDate>Mon, 10 Aug 2026 15:21:32 +0000</pubDate><link>https://news.ycombinator.com/item?id=49244884</link><dc:creator>seanmcdirmid</dc:creator><comments>https://news.ycombinator.com/item?id=49244884</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=49244884</guid></item><item><title><![CDATA[New comment by seanmcdirmid in "Two wheels, a few tradeoffs, and gas prices"]]></title><description><![CDATA[
<p>> You can die unexpectedly from things far less dangerous than motorcycles.<p>Existantials are always trivially true. But statistically:<p>> The lifetime odds of dying as a motorcyclist are roughly 1 in 726. While data specifically tracking "slipping on ice while walking" is rarely isolated on its own, it falls under the broader National Safety Council category for deaths by falls, which has a lifetime odds of 1 in 89.<p>However, if we compare reduced life expectancy:<p>> General lifetime falls reduce life expectancy significantly more across the whole population than motorcycle riding, but a motorcycle crash reduces a single rider's life expectancy far more severely.<p>> Riding a motorcycle poses a vastly higher immediate threat to your lifespan than the everyday risk of falling. On a per-hour basis, riding a motorcycle costs roughly 5 microlives per hour of operation (shaving 2.5 hours off your life expectancy for every hour you ride) because of high speeds and a lack of external frame protection. In contrast, the risk of general falling—despite including hazards like ladders, stairs, and slippery surfaces—costs the average person less than a small fraction of a single microlife per hour of daily movement. While falls ultimately claim more total lives because everyone is exposed to gravity daily, an individual actively trading riding time for walking time dramatically improves their statistical life expectancy by drastically lowering their microlife expenditure.</p>
]]></description><pubDate>Mon, 10 Aug 2026 07:00:19 +0000</pubDate><link>https://news.ycombinator.com/item?id=49240262</link><dc:creator>seanmcdirmid</dc:creator><comments>https://news.ycombinator.com/item?id=49240262</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=49240262</guid></item><item><title><![CDATA[New comment by seanmcdirmid in "The Philippines' big offshoring industry is growing despite AI"]]></title><description><![CDATA[
<p>Has anything really changed for transparency?</p>
]]></description><pubDate>Mon, 10 Aug 2026 06:39:32 +0000</pubDate><link>https://news.ycombinator.com/item?id=49240070</link><dc:creator>seanmcdirmid</dc:creator><comments>https://news.ycombinator.com/item?id=49240070</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=49240070</guid></item><item><title><![CDATA[New comment by seanmcdirmid in "The climate benefits of retiring an operational combustion engine vehicle"]]></title><description><![CDATA[
<p>If you were living in Nepal and your country was getting all of its refined gasoline from India, and therefore that was super expensive, while China was helping your country build a bunch of hydro plants, oh, and they were sending a bunch of EVs to buy over the pass from Tibet, I think your math might change (but then you would highly likely not have that ICE car to begin with).</p>
]]></description><pubDate>Mon, 10 Aug 2026 04:17:08 +0000</pubDate><link>https://news.ycombinator.com/item?id=49239157</link><dc:creator>seanmcdirmid</dc:creator><comments>https://news.ycombinator.com/item?id=49239157</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=49239157</guid></item></channel></rss>