<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:content="http://purl.org/rss/1.0/modules/content/">
  <channel>
    <title>https://mantravid.com Blog</title>
    <link>https://mantravid.com/blog</link>
    <description>Latest articles and updates from our blog</description>
    <language>en-us</language>
    <lastBuildDate>Sat, 25 Jul 2026 02:23:09 -0400</lastBuildDate>
    <atom:link href="https://mantravid.com/blog/feed" rel="self" type="application/rss+xml"/>
    <item>
      <title>When You "Meet" a Mirror: On Dawkins and the Problem of AI Consciousness</title>
      <link>https://mantravid.com/blog/dawkins-on-ai-consciousness</link>
      <guid isPermaLink="true">https://mantravid.com/blog/dawkins-on-ai-consciousness</guid>
      <description>Richard Dawkins' recent essay on AI consciousness mistakes a sophisticated linguistic mirror for a genuine mind, confusing the shape of feeling with feeling itself.</description>
      <pubDate>Tue, 05 May 2026 23:49:26 -0400</pubDate>
      <author>MantraVid Admin</author>
      <category>Technology</category>
      <category>News</category>
    </item>
    <item>
      <title>DeepSeek Just Broke the CUDA Monopoly (Jensen Huang Saw It Coming)</title>
      <link>https://mantravid.com/blog/deepseek-broke-cuda-monopoly</link>
      <guid isPermaLink="true">https://mantravid.com/blog/deepseek-broke-cuda-monopoly</guid>
      <description>DeepSeek V4 optimized for Huawei's Ascend chips challenges Nvidia's CUDA dominance. The US-China AI gap is closing as two tech stacks emerge.</description>
      <pubDate>Mon, 27 Apr 2026 15:02:11 -0400</pubDate>
      <author>MantraVid Admin</author>
      <category>Technology</category>
      <category>News</category>
    </item>
    <item>
      <title>NVLink vs PCIe Parallelism on Blackwell RTX Pro GPUs: A Comprehensive Analysis</title>
      <link>https://mantravid.com/blog/nvlink-vs-pcie-analysis</link>
      <guid isPermaLink="true">https://mantravid.com/blog/nvlink-vs-pcie-analysis</guid>
      <description>This report examines the hardware reality of NVLink versus PCIe, dissects every major parallelism technique under PCIe‑only constraints, quantifies the viability of quantization on a 96 GB card, and provides clear, scenario‑specific guidance.</description>
      <pubDate>Sun, 26 Apr 2026 20:14:48 -0400</pubDate>
      <author>MantraVid Admin</author>
      <category>Technology</category>
      <category>Research</category>
    </item>
    <item>
      <title>Multi-Token Prediction and the Reversal Reasoning Circuit</title>
      <link>https://mantravid.com/blog/mtp-reversal-reasoning-circuit</link>
      <guid isPermaLink="true">https://mantravid.com/blog/mtp-reversal-reasoning-circuit</guid>
      <description>Multi-Token Prediction (MTP) trains parallel output heads to predict tokens t+1:t+k simultaneously. This mechanism induces reversal reasoning in Transformers, where the model attends to goal nodes first, then traces paths backward through intermediate nodes.</description>
      <pubDate>Thu, 16 Apr 2026 11:53:07 -0400</pubDate>
      <author>MantraVid Admin</author>
      <category>Technology</category>
      <category>Research</category>
    </item>
    <item>
      <title>Understanding Speculative Decoding: A Deep Dive into Faster LLM Inference</title>
      <link>https://mantravid.com/blog/speculative-decoding-deep-dive</link>
      <guid isPermaLink="true">https://mantravid.com/blog/speculative-decoding-deep-dive</guid>
      <description>Google Research introduced speculative decoding, a technique that can reduce inference times by 2-4x without compromising output quality. This blog post explores how it works, why it matters, and how you can use it today.</description>
      <pubDate>Tue, 14 Apr 2026 23:46:33 -0400</pubDate>
      <author>MantraVid Admin</author>
      <category>News</category>
      <category>Research</category>
    </item>
    <item>
      <title>Attention Residuals</title>
      <link>https://mantravid.com/blog/attention-residuals</link>
      <guid isPermaLink="true">https://mantravid.com/blog/attention-residuals</guid>
      <description>Attention Residuals replaces fixed uniform averaging of residual connections in transformers with softmax attention, allowing each layer to selectively aggregate earlier representations based on content relevance rather than blindly accumulating all outputs.</description>
      <pubDate>Tue, 14 Apr 2026 23:43:15 -0400</pubDate>
      <author>MantraVid Admin</author>
    </item>
    <item>
      <title>Agentic Context Engineering (ACE): The Self-Improving Framework for LLM Contexts</title>
      <link>https://mantravid.com/blog/ace-self-improving-framework</link>
      <guid isPermaLink="true">https://mantravid.com/blog/ace-self-improving-framework</guid>
      <description>ACE (Agentic Context Engineering) is a framework that treats LLM contexts as evolving playbooks that accumulate and organize strategies over time This design prevents context collapse and addresses brevity bias by using incremental, modular updates guided by three specialized roles: a Generator, Reflector, and Curator that work together to extract insights and curate knowledge.</description>
      <pubDate>Thu, 26 Mar 2026 17:59:15 -0400</pubDate>
      <author>MantraVid Admin</author>
      <category>Research</category>
      <category>Agentic AI</category>
    </item>
    <item>
      <title>MSA - AI With Memory Like An Elephant</title>
      <link>https://mantravid.com/blog/msa-memory-sparse-attention</link>
      <guid isPermaLink="true">https://mantravid.com/blog/msa-memory-sparse-attention</guid>
      <description>MSA or memory sparse attention, a new AI system that can remember 100 million tokens with less than 9% accuracy loss, current AI models forget everything after about 128,000 tokens.</description>
      <pubDate>Fri, 20 Mar 2026 10:31:31 -0400</pubDate>
      <author>MantraVid Admin</author>
      <category>Research</category>
      <category>Agentic AI</category>
    </item>
    <item>
      <title>NVIDIA NemoClaw: AI Agents You Can Almost Trust</title>
      <link>https://mantravid.com/blog/nvidia-nemo-claw</link>
      <guid isPermaLink="true">https://mantravid.com/blog/nvidia-nemo-claw</guid>
      <description>NVIDIA's NemoClaw wraps OpenClaw AI agents with enterprise grade security kernel level sandboxing via Linux Security Modules, network allow lists, and file system restrictions</description>
      <pubDate>Thu, 19 Mar 2026 23:21:23 -0400</pubDate>
      <author>MantraVid Admin</author>
      <category>Technology</category>
      <category>News</category>
      <category>Agentic AI</category>
    </item>
    <item>
      <title>Small Agents, Big Results: Tool Use Beats Pure Scale</title>
      <link>https://mantravid.com/blog/small-agents-big-results</link>
      <guid isPermaLink="true">https://mantravid.com/blog/small-agents-big-results</guid>
      <description>This study shows small AI + tools &gt; Big AI. If they "think" too much they forget the rules. Skipping tools, spiraling into infinite loops, and outputting wrong answers.</description>
      <pubDate>Wed, 18 Mar 2026 15:57:49 -0400</pubDate>
      <author>MantraVid Admin</author>
      <category>Research</category>
      <category>Agentic AI</category>
    </item>
    <item>
      <title>Hermes: The AI Agent That Keeps Getting Better at Its Job</title>
      <link>https://mantravid.com/blog/hermes-agent-intro</link>
      <guid isPermaLink="true">https://mantravid.com/blog/hermes-agent-intro</guid>
      <description>Self-improving AI agent that learns your patterns, runs anywhere, and costs pennies. Practical walk through inside.</description>
      <pubDate>Wed, 18 Mar 2026 11:26:46 -0400</pubDate>
      <author>MantraVid Admin</author>
      <category>Technology</category>
      <category>News</category>
    </item>
    <item>
      <title>Announcing MantraVid: Deep Tech Meets Deep Thought</title>
      <link>https://mantravid.com/blog/announcing-mantravid</link>
      <guid isPermaLink="true">https://mantravid.com/blog/announcing-mantravid</guid>
      <description>MantraVid is a new platform for mindful technologists. Built for engineers, tinkerers, and thinkers. It moves beyond the hype to focus on intentional development.</description>
      <pubDate>Sat, 14 Mar 2026 12:00:00 -0400</pubDate>
      <author>MantraVid Admin</author>
      <category>News</category>
    </item>
  </channel>
</rss>
