<?xml version="1.0" encoding="UTF-8"?>
<rss  xmlns:atom="http://www.w3.org/2005/Atom" 
      xmlns:media="http://search.yahoo.com/mrss/" 
      xmlns:content="http://purl.org/rss/1.0/modules/content/" 
      xmlns:dc="http://purl.org/dc/elements/1.1/" 
      version="2.0">
<channel>
<title>ToKnow.ai</title>
<link>https://toknow.ai/</link>
<atom:link href="https://toknow.ai/index.xml" rel="self" type="application/rss+xml"/>
<description>// Research, Numbers and Insightful Ideas //</description>
<image>
<url>https://toknow.ai/images/logo.png</url>
<title>ToKnow.ai</title>
<link>https://toknow.ai/</link>
<height>144</height>
<width>144</width>
</image>
<generator>quarto-1.6.42</generator>
<lastBuildDate>Mon, 15 Jun 2026 00:00:00 GMT</lastBuildDate>
<item>
  <title>AI Skill: Launch a Production-Grade Store or Blog on Cloudflare, for Free</title>
  <dc:creator>Kabui, Charles</dc:creator>
  <link>https://toknow.ai/posts/build-cloudflare-nextjs-website-ai-skill/</link>
  <description><![CDATA[ A new open-source agent skill launches a production-grade online store, blog, or checkout app on Cloudflare and runs on the free tier: 100,000 requests and 5 million database reads a day, with free file storage and security built in. It asks questions first, then builds in tested phases. ]]></description>
  <category>software-engineering</category>
  <category>artificial-intelligence</category>
  <guid>https://toknow.ai/posts/build-cloudflare-nextjs-website-ai-skill/</guid>
  <pubDate>Mon, 15 Jun 2026 00:00:00 GMT</pubDate>
  <media:content url="https://toknow.ai/posts/build-cloudflare-nextjs-website-ai-skill/image.png" medium="image" type="image/png" height="76" width="144"/>
</item>
<item>
  <title>EU AI Act: Chatbots Must Tell You They Are Not Human from August 2</title>
  <dc:creator>Kabui, Charles</dc:creator>
  <link>https://toknow.ai/posts/eu-ai-act-chatbot-transparency-rules-august-2026/</link>
  <description><![CDATA[ From August 2, 2026, the EU AI Act requires every chatbot, voice assistant, and AI agent to tell people they are talking to a machine, with fines reaching €15 million or 3% of global revenue for companies that ignore it. ]]></description>
  <category>artificial-intelligence</category>
  <guid>https://toknow.ai/posts/eu-ai-act-chatbot-transparency-rules-august-2026/</guid>
  <pubDate>Mon, 15 Jun 2026 00:00:00 GMT</pubDate>
  <media:content url="https://toknow.ai/posts/eu-ai-act-chatbot-transparency-rules-august-2026/image.png" medium="image" type="image/png" height="76" width="144"/>
</item>
<item>
  <title>LatentSkill: Storing Agent Skills as Model Weights, Not Prompt Text</title>
  <dc:creator>Kabui, Charles</dc:creator>
  <link>https://toknow.ai/posts/latentskill-agent-skills-lora-adapters-weight-space/</link>
  <description><![CDATA[ A new method moves an AI agent’s skills out of the prompt and into the model’s weights, cutting skill tokens by up to 72% while raising task success by 21.4% on the ALFWorld benchmark. ]]></description>
  <category>artificial-intelligence</category>
  <guid>https://toknow.ai/posts/latentskill-agent-skills-lora-adapters-weight-space/</guid>
  <pubDate>Mon, 15 Jun 2026 00:00:00 GMT</pubDate>
  <media:content url="https://toknow.ai/posts/latentskill-agent-skills-lora-adapters-weight-space/image.png" medium="image" type="image/png" height="76" width="144"/>
</item>
<item>
  <title>OpenAI Dreaming V3: ChatGPT Now Builds and Updates Its Own Memory of You</title>
  <dc:creator>Kabui, Charles</dc:creator>
  <link>https://toknow.ai/posts/openai-dreaming-v3-chatgpt-self-updating-memory/</link>
  <description><![CDATA[ ChatGPT’s new Dreaming V3 reads your past chats and writes its own profile of you, then keeps it current as things change. In OpenAI’s own tests it recalled personal facts 82.8% of the time, and a 5x cut in serving cost brings it to free users. ]]></description>
  <category>artificial-intelligence</category>
  <guid>https://toknow.ai/posts/openai-dreaming-v3-chatgpt-self-updating-memory/</guid>
  <pubDate>Mon, 15 Jun 2026 00:00:00 GMT</pubDate>
  <media:content url="https://toknow.ai/posts/openai-dreaming-v3-chatgpt-self-updating-memory/image.png" medium="image" type="image/png" height="76" width="144"/>
</item>
<item>
  <title>PaddleOCR-VL-1.6: A 0.9B Document Parser That Beats Models Hundreds of Times Its Size</title>
  <dc:creator>Kabui, Charles</dc:creator>
  <link>https://toknow.ai/posts/paddleocr-vl-16-document-parsing-under-optimized-regions/</link>
  <description><![CDATA[ Baidu’s compact 0.9B model scores 96.33% on OmniDocBench v1.6, topping Gemini 3 Pro and a 235B-parameter rival, by fixing its own weak spots instead of adding more data. ]]></description>
  <category>artificial-intelligence</category>
  <guid>https://toknow.ai/posts/paddleocr-vl-16-document-parsing-under-optimized-regions/</guid>
  <pubDate>Mon, 15 Jun 2026 00:00:00 GMT</pubDate>
  <media:content url="https://toknow.ai/posts/paddleocr-vl-16-document-parsing-under-optimized-regions/image.png" medium="image" type="image/png" height="76" width="144"/>
</item>
<item>
  <title>Reve 2.0: An Image Model That Builds a Layout First, Then Draws</title>
  <dc:creator>Kabui, Charles</dc:creator>
  <link>https://toknow.ai/posts/reve-2-layout-first-image-generation-4k/</link>
  <description><![CDATA[ Reve 2.0 builds every image as an editable layout of positioned elements before rendering, so you can move or rewrite any part afterward. It reached #2 on a public image-quality leaderboard while training on 10x fewer GPUs. ]]></description>
  <category>artificial-intelligence</category>
  <guid>https://toknow.ai/posts/reve-2-layout-first-image-generation-4k/</guid>
  <pubDate>Mon, 15 Jun 2026 00:00:00 GMT</pubDate>
  <media:content url="https://toknow.ai/posts/reve-2-layout-first-image-generation-4k/image.png" medium="image" type="image/png" height="76" width="144"/>
</item>
<item>
  <title>Claude Now Matches the Specialized Software Chemists Use to Identify Molecules</title>
  <dc:creator>Kabui, Charles</dc:creator>
  <link>https://toknow.ai/posts/anthropic-claude-chemist-nmr-spectroscopy/</link>
  <description><![CDATA[ Claude Opus 4.7, a general model with no chemistry training, matched or beat the specialized software chemists use to read a molecule’s magnetic-resonance fingerprint, averaging ±0.079 ppm of error and naming unknown molecules from that data alone. ]]></description>
  <category>artificial-intelligence</category>
  <guid>https://toknow.ai/posts/anthropic-claude-chemist-nmr-spectroscopy/</guid>
  <pubDate>Fri, 12 Jun 2026 00:00:00 GMT</pubDate>
  <media:content url="https://toknow.ai/posts/anthropic-claude-chemist-nmr-spectroscopy/image.png" medium="image" type="image/png" height="76" width="144"/>
</item>
<item>
  <title>Poolside Releases Laguna: Open Coding Models That Run on One GPU</title>
  <dc:creator>Kabui, Charles</dc:creator>
  <link>https://toknow.ai/posts/poolside-laguna-xs2-m1-open-weight-coding-models/</link>
  <description><![CDATA[ Poolside released its first public AI models. The smaller Laguna XS.2 ships with open weights under Apache 2.0, runs on a single GPU, and resolves 68.2% of tasks on the SWE-bench Verified coding benchmark. ]]></description>
  <category>artificial-intelligence</category>
  <category>software-engineering</category>
  <guid>https://toknow.ai/posts/poolside-laguna-xs2-m1-open-weight-coding-models/</guid>
  <pubDate>Fri, 12 Jun 2026 00:00:00 GMT</pubDate>
  <media:content url="https://toknow.ai/posts/poolside-laguna-xs2-m1-open-weight-coding-models/image.png" medium="image" type="image/png" height="76" width="144"/>
</item>
<item>
  <title>RuView: Sensing People With WiFi, Not Cameras</title>
  <dc:creator>Kabui, Charles</dc:creator>
  <link>https://toknow.ai/posts/ruview-wifi-sensing-breathing-pose-no-camera/</link>
  <description><![CDATA[ RuView turns the WiFi signal from a $9 chip into a camera-free people sensor: presence, breathing, heart rate, and 17-point body pose. Its pose model scores 82.69% on a public benchmark, ahead of prior published work. ]]></description>
  <category>artificial-intelligence</category>
  <category>software-engineering</category>
  <guid>https://toknow.ai/posts/ruview-wifi-sensing-breathing-pose-no-camera/</guid>
  <pubDate>Fri, 12 Jun 2026 00:00:00 GMT</pubDate>
  <media:content url="https://toknow.ai/posts/ruview-wifi-sensing-breathing-pose-no-camera/image.png" medium="image" type="image/png" height="76" width="144"/>
</item>
<item>
  <title>Understand-Anything: Turn Any Codebase Into an Interactive Map That Teaches You How It Works</title>
  <dc:creator>Kabui, Charles</dc:creator>
  <link>https://toknow.ai/posts/understand-anything-codebase-knowledge-graph/</link>
  <description><![CDATA[ An open-source plugin sends five AI agents through any codebase and turns it into an interactive map of every file, function, and dependency, so you can see how the whole thing fits together instead of reading it line by line. ]]></description>
  <category>software-engineering</category>
  <category>artificial-intelligence</category>
  <guid>https://toknow.ai/posts/understand-anything-codebase-knowledge-graph/</guid>
  <pubDate>Fri, 12 Jun 2026 00:00:00 GMT</pubDate>
  <media:content url="https://toknow.ai/posts/understand-anything-codebase-knowledge-graph/image.png" medium="image" type="image/png" height="76" width="144"/>
</item>
<item>
  <title>AI This Week: Mythos-Class Claude, a Rebuilt Siri, and OpenAI’s IPO Filing</title>
  <dc:creator>Kabui, Charles</dc:creator>
  <link>https://toknow.ai/posts/ai-weekly-claude-fable-5-apple-siri-ai-openai-ipo/</link>
  <description><![CDATA[ Anthropic’s new Mythos-class Claude leads nearly every benchmark, Apple rebuilds Siri from the ground up, Google translates live speech across 70+ languages, and OpenAI files a confidential S-1 to go public. ]]></description>
  <category>artificial-intelligence</category>
  <guid>https://toknow.ai/posts/ai-weekly-claude-fable-5-apple-siri-ai-openai-ipo/</guid>
  <pubDate>Wed, 10 Jun 2026 00:00:00 GMT</pubDate>
  <media:content url="https://toknow.ai/posts/ai-weekly-claude-fable-5-apple-siri-ai-openai-ipo/image.png" medium="image" type="image/png" height="76" width="144"/>
</item>
<item>
  <title>NVIDIA Cosmos 3: One Open Model for Text, Images, Video, Audio, and Robot Actions</title>
  <dc:creator>Kabui, Charles</dc:creator>
  <link>https://toknow.ai/posts/nvidia-cosmos-3-omnimodal-world-model-physical-ai/</link>
  <description><![CDATA[ NVIDIA’s Cosmos 3 is an open family of world models that handle text, images, video, audio, and robot actions in one system, in 16B and 64B sizes. NVIDIA reports it tops open-source rankings for image generation, video generation, and robot control. ]]></description>
  <category>artificial-intelligence</category>
  <guid>https://toknow.ai/posts/nvidia-cosmos-3-omnimodal-world-model-physical-ai/</guid>
  <pubDate>Wed, 10 Jun 2026 00:00:00 GMT</pubDate>
  <media:content url="https://toknow.ai/posts/nvidia-cosmos-3-omnimodal-world-model-physical-ai/image.png" medium="image" type="image/png" height="76" width="144"/>
</item>
<item>
  <title>Heretic: A Python Tool That Automatically Strips Safety Guardrails from Any Language Model</title>
  <dc:creator>Kabui, Charles</dc:creator>
  <link>https://toknow.ai/posts/heretic-automatic-censorship-removal-language-models-abliteration/</link>
  <description><![CDATA[ Heretic uses directional ablation to remove alignment training from transformer models fully automatically. Over 4,000 community-created models on HuggingFace so far, with lower quality loss than manual approaches. ]]></description>
  <category>artificial-intelligence</category>
  <guid>https://toknow.ai/posts/heretic-automatic-censorship-removal-language-models-abliteration/</guid>
  <pubDate>Sat, 06 Jun 2026 00:00:00 GMT</pubDate>
  <media:content url="https://toknow.ai/posts/heretic-automatic-censorship-removal-language-models-abliteration/image.png" medium="image" type="image/png" height="76" width="144"/>
</item>
<item>
  <title>OpenAI’s Tax AI Went from 25% to 86% Accuracy in Six Weeks by Fixing Its Own Mistakes</title>
  <dc:creator>Kabui, Charles</dc:creator>
  <link>https://toknow.ai/posts/openai-codex-self-improving-tax-ai-25-to-86-accuracy/</link>
  <description><![CDATA[ OpenAI and Thrive Holdings built a tax preparation AI that uses Codex to investigate its own errors, write fixes, and redeploy, processing 7,000 returns across 30+ accounting firms with 97% field accuracy. ]]></description>
  <category>artificial-intelligence</category>
  <guid>https://toknow.ai/posts/openai-codex-self-improving-tax-ai-25-to-86-accuracy/</guid>
  <pubDate>Sat, 06 Jun 2026 00:00:00 GMT</pubDate>
  <media:content url="https://toknow.ai/posts/openai-codex-self-improving-tax-ai-25-to-86-accuracy/image.png" medium="image" type="image/png" height="76" width="144"/>
</item>
<item>
  <title>VoxCPM2: Tokenizer-Free TTS Generates 48kHz Speech in 30 Languages Without an Audio Codec</title>
  <dc:creator>Kabui, Charles</dc:creator>
  <link>https://toknow.ai/posts/voxcpm2-tokenizer-free-tts-30-languages-voice-design/</link>
  <description><![CDATA[ Tsinghua’s 2B-parameter VoxCPM2 skips discrete audio tokens entirely, generating studio-quality 48kHz speech directly from text in 30 languages with voice cloning from a short clip. ]]></description>
  <category>artificial-intelligence</category>
  <guid>https://toknow.ai/posts/voxcpm2-tokenizer-free-tts-30-languages-voice-design/</guid>
  <pubDate>Sat, 06 Jun 2026 00:00:00 GMT</pubDate>
  <media:content url="https://toknow.ai/posts/voxcpm2-tokenizer-free-tts-30-languages-voice-design/image.png" medium="image" type="image/png" height="76" width="144"/>
</item>
<item>
  <title>YOLOv12 Pill Counting: Attention-Based Object Detection Meets Pharmacy Automation</title>
  <dc:creator>Kabui, Charles</dc:creator>
  <link>https://toknow.ai/posts/yolov12-pill-counting-attention-centric-real-time-object-detection/</link>
  <description><![CDATA[ YOLOv12 brings attention mechanisms to real-time object detection, achieving 40.6% mAP at 1.64ms latency. A new open tutorial shows how to fine-tune it for automated pill counting in pharmacies. ]]></description>
  <category>artificial-intelligence</category>
  <guid>https://toknow.ai/posts/yolov12-pill-counting-attention-centric-real-time-object-detection/</guid>
  <pubDate>Sat, 06 Jun 2026 00:00:00 GMT</pubDate>
  <media:content url="https://toknow.ai/posts/yolov12-pill-counting-attention-centric-real-time-object-detection/image.png" medium="image" type="image/png" height="76" width="144"/>
</item>
<item>
  <title>MathWorks Open-Sources a Complete Robotics Playbook for MATLAB and Simulink</title>
  <dc:creator>Kabui, Charles</dc:creator>
  <link>https://toknow.ai/posts/mathworks-awesome-matlab-robotics-open-source-ros2-simulink/</link>
  <description><![CDATA[ A curated collection of ready-to-run demos spans 11 robotics domains, from off-road haul trucks to underwater vehicles, with full ROS 2 integration and Simulink-to-hardware deployment across 32 open repositories. ]]></description>
  <category>software-engineering</category>
  <guid>https://toknow.ai/posts/mathworks-awesome-matlab-robotics-open-source-ros2-simulink/</guid>
  <pubDate>Fri, 05 Jun 2026 00:00:00 GMT</pubDate>
  <media:content url="https://toknow.ai/posts/mathworks-awesome-matlab-robotics-open-source-ros2-simulink/image.png" medium="image" type="image/png" height="76" width="144"/>
</item>
<item>
  <title>RecursiveMAS: Multi-Agent Systems That Think in Latent Space Instead of Text</title>
  <dc:creator>Kabui, Charles</dc:creator>
  <link>https://toknow.ai/posts/recursive-multi-agent-systems-latent-space-recursion-uiuc-stanford-nvidia-mit/</link>
  <description><![CDATA[ A framework from UIUC, Stanford, NVIDIA, and MIT replaces text-based agent communication with latent-state transfer, delivering +8.3% accuracy, 2.4x speedup, and 75.6% fewer tokens across 9 benchmarks. ]]></description>
  <category>artificial-intelligence</category>
  <guid>https://toknow.ai/posts/recursive-multi-agent-systems-latent-space-recursion-uiuc-stanford-nvidia-mit/</guid>
  <pubDate>Fri, 05 Jun 2026 00:00:00 GMT</pubDate>
  <media:content url="https://toknow.ai/posts/recursive-multi-agent-systems-latent-space-recursion-uiuc-stanford-nvidia-mit/image.png" medium="image" type="image/png" height="76" width="144"/>
</item>
<item>
  <title>AI Memory Problem: Why Bigger Context Windows Don’t Stop Agents From Forgetting</title>
  <dc:creator>Kabui, Charles</dc:creator>
  <link>https://toknow.ai/posts/ai-memory-problem-context-window-persistent-agent-memory/</link>
  <description><![CDATA[ Three open-source memory engines tackle the AI context problem, with Supermemory scoring 81.6% on LongMemEval and Agentmemory cutting token usage by 92% for coding agents. ]]></description>
  <category>artificial-intelligence</category>
  <guid>https://toknow.ai/posts/ai-memory-problem-context-window-persistent-agent-memory/</guid>
  <pubDate>Wed, 03 Jun 2026 00:00:00 GMT</pubDate>
  <media:content url="https://toknow.ai/posts/ai-memory-problem-context-window-persistent-agent-memory/image.png" medium="image" type="image/png" height="76" width="144"/>
</item>
<item>
  <title>Anthropic Raises $65B at $965B Valuation, Surpassing OpenAI</title>
  <dc:creator>Kabui, Charles</dc:creator>
  <link>https://toknow.ai/posts/anthropic-65b-series-h-965b-valuation-surpassing-openai/</link>
  <description><![CDATA[ Anthropic’s Series H round values the company at $965 billion, surpassing OpenAI’s $852 billion. Run-rate revenue has crossed $47 billion, and the company has filed for an IPO. ]]></description>
  <category>artificial-intelligence</category>
  <category>entrepreneurship</category>
  <guid>https://toknow.ai/posts/anthropic-65b-series-h-965b-valuation-surpassing-openai/</guid>
  <pubDate>Wed, 03 Jun 2026 00:00:00 GMT</pubDate>
  <media:content url="https://toknow.ai/posts/anthropic-65b-series-h-965b-valuation-surpassing-openai/image.png" medium="image" type="image/png" height="76" width="144"/>
</item>
</channel>
</rss>
