<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: vexna</title><link>https://news.ycombinator.com/user?id=vexna</link><description>Hacker News RSS</description><docs>https://hnrss.org/</docs><generator>hnrss v2.1.1</generator><lastBuildDate>Sun, 26 Apr 2026 10:23:56 +0000</lastBuildDate><atom:link href="https://hnrss.org/user?id=vexna" rel="self" type="application/rss+xml"></atom:link><item><title><![CDATA[New comment by vexna in "GPT-5.5"]]></title><description><![CDATA[
<p>There's an asterisk right below that table stating that:<p>> *Anthropic reported  signs of memorization on a subset of problems<p>And from the Anthropic's Opus 4.7 release page, it also states:<p>> SWE-bench Verified, Pro, and Multilingual: Our memorization screens flag a subset of problems in these SWE-bench evals. Excluding any problems that show signs of memorization, Opus 4.7’s margin of improvement over Opus 4.6 holds.</p>
]]></description><pubDate>Thu, 23 Apr 2026 18:31:38 +0000</pubDate><link>https://news.ycombinator.com/item?id=47879588</link><dc:creator>vexna</dc:creator><comments>https://news.ycombinator.com/item?id=47879588</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=47879588</guid></item><item><title><![CDATA[New comment by vexna in "OpenCode – Open source AI coding agent"]]></title><description><![CDATA[
<p>They are using effect-ts which uses yield as their equivalent to haskells Do notation.</p>
]]></description><pubDate>Sat, 21 Mar 2026 13:21:24 +0000</pubDate><link>https://news.ycombinator.com/item?id=47466778</link><dc:creator>vexna</dc:creator><comments>https://news.ycombinator.com/item?id=47466778</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=47466778</guid></item><item><title><![CDATA[New comment by vexna in "Claude Code gets native LSP support"]]></title><description><![CDATA[
<p>Just a heads up that this is completely broken as of 2.0.76.<p>Dug through their obfuscated JS and it looks like they forgot to re-add a function call in the LSP manager initialize function that actually adds/indexes the lsp registered from plugins.</p>
]]></description><pubDate>Tue, 23 Dec 2025 03:44:01 +0000</pubDate><link>https://news.ycombinator.com/item?id=46362211</link><dc:creator>vexna</dc:creator><comments>https://news.ycombinator.com/item?id=46362211</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=46362211</guid></item><item><title><![CDATA[New comment by vexna in "Claude’s memory architecture is the opposite of ChatGPT’s"]]></title><description><![CDATA[
<p>No real modulation or switching occurs.
If you start a new chat, your “explicit” memories will pretty much be injected right into the system prompt (I almost think of it as compile time memory). The other memories can sort of thought of as “runtime” memory: your message will be queried against the embeddings of your chat memories and if a strong match is made, the model will use the embedding data it matches against.</p>
]]></description><pubDate>Sun, 14 Sep 2025 18:23:36 +0000</pubDate><link>https://news.ycombinator.com/item?id=45242073</link><dc:creator>vexna</dc:creator><comments>https://news.ycombinator.com/item?id=45242073</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=45242073</guid></item><item><title><![CDATA[New comment by vexna in "Claude’s memory architecture is the opposite of ChatGPT’s"]]></title><description><![CDATA[
<p>ChatGPT has 2 types of memory: The “explicit” memory you tell it to remember (sometimes triggers when it thinks you say something important) and the global/project level automated memory that are stored as embeddings.<p>The explicit memory is what you see in the memory section of the UI and is pretty much injected directly into the system prompt.<p>The global embeddings memory is accessed via runtime vector search.<p>Sadly I wish I could disable the embeddings memory and keep the explicit. The lossy nature of embeddings make it hallucinate a bit too much for my liking and GPT-5 seems to have just made it worse.</p>
]]></description><pubDate>Fri, 12 Sep 2025 08:40:15 +0000</pubDate><link>https://news.ycombinator.com/item?id=45220045</link><dc:creator>vexna</dc:creator><comments>https://news.ycombinator.com/item?id=45220045</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=45220045</guid></item><item><title><![CDATA[New comment by vexna in "From GPT-4 to GPT-5: Measuring progress through MedHELM [pdf]"]]></title><description><![CDATA[
<p>kool aid or not -- "reasoning" is already part of the LLM verbiage (e.g `reasoning` models having `reasoningBudget`).
The meaning might not be 1:1 to human reasoning, but when the LLM shows its "reasoning" it does look _appear_ like a train of thought. If I had to give what it's doing a name (like I'm naming a function), I'd be hard pressed to not go with something like `reason`/`think`.</p>
]]></description><pubDate>Fri, 22 Aug 2025 11:13:51 +0000</pubDate><link>https://news.ycombinator.com/item?id=44983096</link><dc:creator>vexna</dc:creator><comments>https://news.ycombinator.com/item?id=44983096</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=44983096</guid></item><item><title><![CDATA[New comment by vexna in "Gemini CLI"]]></title><description><![CDATA[
<p>It gets even more confusing! If you're on the "Premium" plans (i.e the the old standard "Google One" plans) and upgrade to >=5TB storage, your "Premium" plan starts including all the features of "Google AI Pro".<p>Tip: If you do annual billing for "Premium (5 TB)", you end up paying $21/month for 5TB of storage and the same AI features of "Google AI pro (2TB)"; which is only $1/month more than doing "Google AI Pro (2 TB)" (which only has monthly billing)</p>
]]></description><pubDate>Wed, 25 Jun 2025 18:44:03 +0000</pubDate><link>https://news.ycombinator.com/item?id=44380595</link><dc:creator>vexna</dc:creator><comments>https://news.ycombinator.com/item?id=44380595</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=44380595</guid></item><item><title><![CDATA[New comment by vexna in "British Columbia to recriminalize use of drugs in public spaces"]]></title><description><![CDATA[
<p>Guessing you mean the McDonalds on Pandora? I remember going there a lot during high school and walked by it again recently while visiting the city for the the first time since 2009. Sketchy doesn't begin to describe it. Was really sad to see what happened to the city I grew up in.</p>
]]></description><pubDate>Sat, 27 Apr 2024 09:14:17 +0000</pubDate><link>https://news.ycombinator.com/item?id=40178546</link><dc:creator>vexna</dc:creator><comments>https://news.ycombinator.com/item?id=40178546</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=40178546</guid></item><item><title><![CDATA[New comment by vexna in ""Fake Chinese income" mortgages fuel Toronto real estate bubble: HSBC bank leaks"]]></title><description><![CDATA[
<p>Canada currently has an issue with handing out way too many student visas to students going to strip mall diploma mills. I believe the federal government has just enforced a cap for the next 2 years.</p>
]]></description><pubDate>Tue, 06 Feb 2024 21:29:12 +0000</pubDate><link>https://news.ycombinator.com/item?id=39280907</link><dc:creator>vexna</dc:creator><comments>https://news.ycombinator.com/item?id=39280907</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=39280907</guid></item><item><title><![CDATA[New comment by vexna in "Google will retire Gmail’s Basic HTML view in January 2024"]]></title><description><![CDATA[
<p>You can reenable the web interface in iCloud settings under system preferences on a Mac (not sure if it’s on iOS). The web interface will require just in time verification with an iCloud device to confirm log ins though. Works pretty well.</p>
]]></description><pubDate>Sun, 24 Sep 2023 19:10:43 +0000</pubDate><link>https://news.ycombinator.com/item?id=37636228</link><dc:creator>vexna</dc:creator><comments>https://news.ycombinator.com/item?id=37636228</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=37636228</guid></item><item><title><![CDATA[New comment by vexna in "Java 20 / JDK 20: General Availability"]]></title><description><![CDATA[
<p>Its incredibly powerful! One downside is variables captured in the match are lifted to the function level. The lack of block level scoping in the match statement leads to a lot of people shooting themselves in the foot by overwriting a variable that they didn't intend to. On a personal note, I find the 2 levels of indentation to reach the case body to be too much as well... but that's just a personal preference.</p>
]]></description><pubDate>Tue, 21 Mar 2023 16:40:02 +0000</pubDate><link>https://news.ycombinator.com/item?id=35248713</link><dc:creator>vexna</dc:creator><comments>https://news.ycombinator.com/item?id=35248713</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=35248713</guid></item><item><title><![CDATA[New comment by vexna in "Bun v0.5"]]></title><description><![CDATA[
<p>Because its for running server side javascript, just like node/deno. It would make sense to compare itself to its competition.</p>
]]></description><pubDate>Wed, 18 Jan 2023 20:55:09 +0000</pubDate><link>https://news.ycombinator.com/item?id=34432945</link><dc:creator>vexna</dc:creator><comments>https://news.ycombinator.com/item?id=34432945</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=34432945</guid></item><item><title><![CDATA[New comment by vexna in "I cut GTA Online loading times (2021)"]]></title><description><![CDATA[
<p>Because the existence of the problem bothers you implicitly. Or because it’s usually a good way to a promo</p>
]]></description><pubDate>Fri, 10 Jun 2022 05:51:45 +0000</pubDate><link>https://news.ycombinator.com/item?id=31690530</link><dc:creator>vexna</dc:creator><comments>https://news.ycombinator.com/item?id=31690530</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=31690530</guid></item><item><title><![CDATA[New comment by vexna in "Bflat: C# as you know it but with Go-like tooling"]]></title><description><![CDATA[
<p>replied with this on another post, but you can add `-p:PublishReadyToRun=true` to your `dotnet publish` command to do AOT. it blows up your binary in size, but pretty much gives your warmed up code from the start.</p>
]]></description><pubDate>Fri, 25 Jun 2021 19:01:57 +0000</pubDate><link>https://news.ycombinator.com/item?id=27634925</link><dc:creator>vexna</dc:creator><comments>https://news.ycombinator.com/item?id=27634925</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=27634925</guid></item><item><title><![CDATA[New comment by vexna in "Bflat: C# as you know it but with Go-like tooling"]]></title><description><![CDATA[
<p>You can actually add `-p:PublishReadyToRun=true` to your `dotnet publish` command and it will do AOT compilation for you. It blows up your executable a bit, but it does pretty much give you warmed up code at the get go.</p>
]]></description><pubDate>Fri, 25 Jun 2021 19:00:19 +0000</pubDate><link>https://news.ycombinator.com/item?id=27634908</link><dc:creator>vexna</dc:creator><comments>https://news.ycombinator.com/item?id=27634908</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=27634908</guid></item><item><title><![CDATA[New comment by vexna in "Deno 1.9"]]></title><description><![CDATA[
<p>I found most of the dependency woes to go away with import maps: <a href="https://deno.land/manual@v1.9.0/linking_to_external_code/import_maps" rel="nofollow">https://deno.land/manual@v1.9.0/linking_to_external_code/imp...</a></p>
]]></description><pubDate>Tue, 13 Apr 2021 22:20:09 +0000</pubDate><link>https://news.ycombinator.com/item?id=26800110</link><dc:creator>vexna</dc:creator><comments>https://news.ycombinator.com/item?id=26800110</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=26800110</guid></item></channel></rss>