<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: namanbnsl</title><link>https://news.ycombinator.com/user?id=namanbnsl</link><description>Hacker News RSS</description><docs>https://hnrss.org/</docs><generator>hnrss v2.1.1</generator><lastBuildDate>Wed, 18 Nov 2026 14:19:46 +0000</lastBuildDate><atom:link href="https://hnrss.org/user?id=namanbnsl" rel="self" type="application/rss+xml"></atom:link><item><title><![CDATA[We rewrote one weight matrix to teach a 0.8B model conversation classification]]></title><description><![CDATA[
<p>Article URL: <a href="https://giga.ai/news/one-weight-matrix">https://giga.ai/news/one-weight-matrix</a></p>
<p>Comments URL: <a href="https://news.ycombinator.com/item?id=50405905">https://news.ycombinator.com/item?id=50405905</a></p>
<p>Points: 1</p>
<p># Comments: 0</p>
]]></description><pubDate>Sat, 07 Nov 2026 16:41:05 +0000</pubDate><link>https://giga.ai/news/one-weight-matrix</link><dc:creator>namanbnsl</dc:creator><comments>https://news.ycombinator.com/item?id=50405905</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50405905</guid></item><item><title><![CDATA[EmbeddingGemma 2: Unified and Efficient Multimodal Representations]]></title><description><![CDATA[
<p>Article URL: <a href="https://arxiv.org/abs/2610.37364">https://arxiv.org/abs/2610.37364</a></p>
<p>Comments URL: <a href="https://news.ycombinator.com/item?id=50391872">https://news.ycombinator.com/item?id=50391872</a></p>
<p>Points: 3</p>
<p># Comments: 0</p>
]]></description><pubDate>Fri, 06 Nov 2026 13:06:35 +0000</pubDate><link>https://arxiv.org/abs/2610.37364</link><dc:creator>namanbnsl</dc:creator><comments>https://news.ycombinator.com/item?id=50391872</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50391872</guid></item><item><title><![CDATA[Gemini Omni 1.2 Flash: Fast and cost-effective with longer clips]]></title><description><![CDATA[
<p>Article URL: <a href="https://blog.google/innovation-and-ai/technology/developers-tools/introducing-gemini-omni-1-2-flash/">https://blog.google/innovation-and-ai/technology/developers-tools/introducing-gemini-omni-1-2-flash/</a></p>
<p>Comments URL: <a href="https://news.ycombinator.com/item?id=50381613">https://news.ycombinator.com/item?id=50381613</a></p>
<p>Points: 6</p>
<p># Comments: 1</p>
]]></description><pubDate>Thu, 05 Nov 2026 17:10:09 +0000</pubDate><link>https://blog.google/innovation-and-ai/technology/developers-tools/introducing-gemini-omni-1-2-flash/</link><dc:creator>namanbnsl</dc:creator><comments>https://news.ycombinator.com/item?id=50381613</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50381613</guid></item><item><title><![CDATA[The Stacks Project, Formalized in Lean]]></title><description><![CDATA[
<p>Article URL: <a href="https://axiommath.ai/research/the-stacks-project-formalized/">https://axiommath.ai/research/the-stacks-project-formalized/</a></p>
<p>Comments URL: <a href="https://news.ycombinator.com/item?id=50380979">https://news.ycombinator.com/item?id=50380979</a></p>
<p>Points: 97</p>
<p># Comments: 26</p>
]]></description><pubDate>Thu, 05 Nov 2026 16:16:59 +0000</pubDate><link>https://axiommath.ai/research/the-stacks-project-formalized/</link><dc:creator>namanbnsl</dc:creator><comments>https://news.ycombinator.com/item?id=50380979</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50380979</guid></item><item><title><![CDATA[Investigating Vault Particles with Claude]]></title><description><![CDATA[
<p>Article URL: <a href="https://www.anthropic.com/research/investigating-vault-particles-with-claude">https://www.anthropic.com/research/investigating-vault-particles-with-claude</a></p>
<p>Comments URL: <a href="https://news.ycombinator.com/item?id=50378181">https://news.ycombinator.com/item?id=50378181</a></p>
<p>Points: 4</p>
<p># Comments: 0</p>
]]></description><pubDate>Thu, 05 Nov 2026 12:08:39 +0000</pubDate><link>https://www.anthropic.com/research/investigating-vault-particles-with-claude</link><dc:creator>namanbnsl</dc:creator><comments>https://news.ycombinator.com/item?id=50378181</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50378181</guid></item><item><title><![CDATA[New comment by namanbnsl in "Ask HN: What are the fundamentals of agentic coding?"]]></title><description><![CDATA[
<p>I think one fundamental for getting agents to work well is giving them a number to hill-climb.<p>If you can create an eval for something in your project, the agent can keep trying different approaches until the score goes up.<p>I believe building good evals will matter more than ever.</p>
]]></description><pubDate>Thu, 05 Nov 2026 12:05:15 +0000</pubDate><link>https://news.ycombinator.com/item?id=50378155</link><dc:creator>namanbnsl</dc:creator><comments>https://news.ycombinator.com/item?id=50378155</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50378155</guid></item><item><title><![CDATA[OpenAI LifeVal]]></title><description><![CDATA[
<p>Article URL: <a href="https://openai.com/index/lifeval/">https://openai.com/index/lifeval/</a></p>
<p>Comments URL: <a href="https://news.ycombinator.com/item?id=50365924">https://news.ycombinator.com/item?id=50365924</a></p>
<p>Points: 1</p>
<p># Comments: 0</p>
]]></description><pubDate>Wed, 04 Nov 2026 14:22:12 +0000</pubDate><link>https://openai.com/index/lifeval/</link><dc:creator>namanbnsl</dc:creator><comments>https://news.ycombinator.com/item?id=50365924</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50365924</guid></item><item><title><![CDATA[New comment by namanbnsl in "Astra can read and write weights to improve Qwen"]]></title><description><![CDATA[
<p>Mainly due to the looping.<p>Before the edit, 93% of MATH-500 answers hit the 32k token limit, compared to 33% after.<p>With 10 and 25 forward passes, looping barely decreased (87% and 86% still hit the limit) and the score didn't change much. The score jump only happened at 50 forward passes when looping decreased.<p>Astra edited the weights to bring out a capability the model already had rather than enhancing the ability itself. Pretty impressive.</p>
]]></description><pubDate>Tue, 03 Nov 2026 11:50:48 +0000</pubDate><link>https://news.ycombinator.com/item?id=50350837</link><dc:creator>namanbnsl</dc:creator><comments>https://news.ycombinator.com/item?id=50350837</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=50350837</guid></item></channel></rss>