[{"channel_id":1634145994,"post_id":1970,"date":1794719431000,"views":"29","text":"More from NVIDIA: a new open-source project - auto ontology, looks super cool 👇🏼<br><br><a target=\"_blank\" rel=\"noreferrer nofollow\" href=\"https://www.youtube.com/watch?v=zyuq3CytzQM\">https://www.youtube.com/watch?v=zyuq3CytzQM</a>","text_length":127,"media":{"root":"/013/sgcAAMoWZ2EAAAAAJC97GOG_yXw","photo":{"thumbs":{"m":{"w":320,"h":180,"hash":"sd8xpGwe1YnoweNYDX27mg&ts=1794905668"},"x":{"w":800,"h":450,"hash":"ljc1iGB4U02w7hX6hE-A1g&ts=1794905668"},"y":{"w":1280,"h":720,"hash":"DMRU_IKt3T3QH5w_8MXuTA&ts=1794905668"},"i":{"bytes":"AXACg|DLjt5pYy6ISoOM/hmpBY3Jz+6PHXkVagLx2gjA4ZWzj3/+sBU7StkuV3b+CD6YI/qannQGYbO4BwYjnBP4CkhgMgznC5xmtQ3UigEjoMdO3p/n0rPOYWxExIAywNLmvsIklskS3aVZSSP4SuKKZNI4R1MaqOMkgg0URvbUCMSqRh4waYwQn5SR7GiiqsA3bjvToyin5s+3HSiigB8jo45JZv7xHNFFFCA//9k="}}}}},{"channel_id":1634145994,"post_id":1968,"date":1794717952000,"forwards":"1","views":"15","text":"NVIDIA&#039;s Nemotron Bolt 74B-A9B 2 Bit 👇🏼👇🏼👇🏼<br><br>The following video by Sam Witteveen explains how NVIDIA combines pruning, distillation, post-training, and quantization-aware training in Nemotron 3.5 Bolt. It also examines the model&#039;s mixed-precision design and tests it on RAG, tool calling, and concurrent inference tasks.<br><br><a target=\"_blank\" rel=\"noreferrer nofollow\" href=\"https://www.youtube.com/watch?v=WJpeW9lCSUM\">https://www.youtube.com/watch?v=WJpeW9lCSUM</a>","text_length":372,"media":{"root":"/011/sAcAAMoWZ2EAAAAAc_VhVqkVMKU","webpage":{"url":"https://youtube.com/watch?v=WJpeW9lCSUM","type":"video","title":"NVIDIA's Nemotron Bolt 74B-A9B 2 Bit","site_name":"YouTube","display_url":"youtube.com/watch?v=WJpeW9lCSUM","description":"In this video, I look at a new model and a set of techniques that NVIDIA has used to make this 2-bit model perform at a level much better than expected. \n\nResearch Report: https://research.nvidia.com/labs/nemotron/Nemotron-3.5-Bolt-24GB/\nHF: https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Bolt-74B-A9B-24GB\nParetoQ: https://arxiv.org/pdf/2502.02631\n\n2nd Channel:  https://www.youtube.com/channel/UCUF7WtR4JpVbQqcQGrLeADw\n\nTwitter: https://x.com/Sam_Witteveen \n\n🕵️ Interested in building LLM Agents? Fill out the form below\nBuilding LLM Agents Form: https://drp.li/dIMes\n\n👨‍💻Github:\nhttps://github.com/samwit/llm-tutorials\n\n⏱️Time Stamps:\n0:00 Intro\n0:14 Microsoft’s secret 2-bit model\n1:02 Nemotron 3.5 Bolt 74B-A9B\n1:54 Why the recipe matters\n3:00 How normal quantization works\n3:43 Why going below 4-bit breaks\n4:38 ParetoQ\n5:31 Why 2-bit matters\n7:04 Mixed precision\n8:02 Stage 1: Pruning\n9:23 Stage 2: Post-training\n9:40 Stage 3: Quantization-aware training\n11:48 NVIDIA’s twist: QAD\n14:29 Demo on the ZGX Fury\n16:00 Pelican SVG test\n17:55 Agentic tool-calling tests\n19:42 Concurrency tests\n20:47 Which hardware Bolt is for\n22:04 Conclusion","thumbs":{"m":{"w":320,"h":180,"hash":"Woi1OGvpKZVbaU-G9BvxNg&ts=1794905668"},"x":{"w":800,"h":450,"hash":"uAr_m3Kh8lBZNRkfMFnm-Q&ts=1794905668"},"y":{"w":1280,"h":720,"hash":"yN6ytsmI9K1M4dJKfx9pfg&ts=1794905668"},"i":{"bytes":"AXACg|CUyKjFTG5xjkDijzU/55yflVWRGFwSYztLg7iM54P+faq5jbAwhGEx75wf61zKnFomyNrMaLuZ1A+tRG+h+barMi9WqjKmy2lkVNqlkxlcZGP8aq+am0gqfoDxVRpKw7GozLcuGhmyByV6HFFV9LYtM/AA2dB9RRWU1Z2RLITeK20uudvSl+1x/wBw5/3jRRXRyIdhz3Ky2rRYwO1UvKypYHpRRRHTYES28vkoxHU96KKKHFMGf//Z"}},"embed":{"url":"https://www.youtube.com/embed/WJpeW9lCSUM","type":"iframe","w":1280,"h":720}}}},{"channel_id":1634145994,"post_id":1967,"date":1794628052000,"forwards":"4","views":"132","text":"Local AI 301: Quantization 🚀<br><br>This Hugging Face masterclass covers low-precision model fundamentals, GGUF and <code>llama.cpp</code> quantization, calibration with importance matrices, and production inference formats in vLLM.<br><br><a target=\"_blank\" rel=\"noreferrer nofollow\" href=\"https://www.youtube.com/watch?v=il0dgxjpoC0\">https://www.youtube.com/watch?v=il0dgxjpoC0</a>","text_length":259,"media":{"root":"/00e/rwcAAMoWZ2EAAAAA7MNun_1I8lw","webpage":{"url":"https://youtube.com/watch?v=il0dgxjpoC0","type":"video","title":"Local AI 301: Quantization","site_name":"YouTube","display_url":"youtube.com/watch?v=il0dgxjpoC0","description":"Learn about quantization in this livestream and get started with quantization using llama.cpp and vLLM.\nProgram: \n-  Introduction to Quantization, Merve Noyan, Hugging Face\n- Calibration based quantization, Colin Kealty (aka bartowski), RedHat\n- Quantization Aware Distillation, Maxime Labonne, Aditya Tadimeti, LiquidAI\n- Fast Microscaling Formats on Any GPU with vLLM, Michael Goin, RedHat","thumbs":{"m":{"w":320,"h":180,"hash":"SmU91baCs-Loma5T0kUy2g&ts=1794905668"},"x":{"w":800,"h":450,"hash":"4t05qHwUlqVaqlWRb3nQjg&ts=1794905668"},"y":{"w":1280,"h":720,"hash":"Yj7tIbeF-ORV6A7qCEdPrA&ts=1794905668"},"i":{"bytes":"AXACg|DQkkdZCFQkeuaPNmKqVjznr7U9z89IG6D+tZcr7lXEWSYyhTFhM/ezUxYAZNC8iq83mFvRB3AzQ7xQbkpljXGWHPQUVTWD7UA/mHb2OP5UVcbNasRaZpPNI8olB/HuH8qXDf3D/wB9UUVMnYByjI5XGPenYGMdqKKaARFVFwqhR6AYooopgf/Z"}},"embed":{"url":"https://www.youtube.com/embed/il0dgxjpoC0","type":"iframe","w":1280,"h":720}}}},{"channel_id":1634145994,"post_id":1966,"date":1794068800000,"views":"124","text":"Issue 113 is out! <br><br>This week&#039;s focus:<br>🔹 Open Source of the Week - Onyxia<br>🔹 New learning resources - Python voice assistants, Ripple-65B Preview, vector database trade-offs, and RAG from scratch<br>🔹 Book of the week - Graph Engineering — The Loop Memory Model by Volodymyr Pavlyshyn<br><br><a target=\"_blank\" rel=\"noreferrer nofollow\" href=\"https://ramikrispin.substack.com/p/graph-engineering-onyxia-ripple-65b\">https://ramikrispin.substack.com/p/graph-engineering-onyxia-ripple-65b</a>","text_length":355,"media":{"root":"/00d/rgcAAMoWZ2EAAAAA6_Fg6skwlLw","webpage":{"url":"https://ramikrispin.substack.com/p/graph-engineering-onyxia-ripple-65b","type":"photo","title":"Graph Engineering, Onyxia, Ripple-65B, and Vector Databases | Issue 113","site_name":"Substack","display_url":"ramikrispin.substack.com/p/graph-engineering-onyxia-ripple-65b","description":"Explore Graph Engineering's approach to durable agent memory, Onyxia's Kubernetes data-science environment, Python voice assistants, Ripple-65B Preview, vector database tradeoffs, and RAG.","author":"Rami Krispin","thumbs":{"m":{"w":320,"h":160,"hash":"aRKVURm85OmGY6AwOq7cKg&ts=1794905668"},"x":{"w":800,"h":400,"hash":"ehsZboRvy7nvB8JGVKiwLA&ts=1794905668"},"y":{"w":1280,"h":640,"hash":"wb_v74IG7yi3j8OhwPJCYw&ts=1794905668"},"w":{"w":1600,"h":800,"hash":"we2gtH7en6WeR-iKlNIR9Q&ts=1794905668"},"i":{"bytes":"AUACg|CEOp68ChNjsSpyOlVpHHCgj3xSxvzn070OTsO2pbG0DOcj1xSB4yeSajin3hkJyR0NNyCMZxRzsORFkNH/AHqKrLgD1/DFFHOL2ZXIA5Apw60UVJY7A9OlFFFAxpHyknniiiigD//Z"}}}}},{"channel_id":1634145994,"post_id":1965,"date":1793766874000,"views":"38","text":"Started a new series that focuses on machine learning algorithms 🚀<br><br>Yes, if you are using ML, you don&#039;t really build the algorithm from scratch and, most likely, will chat with your coding agent and ask it to build it. <br><br>But what should you ask?<br><br>The goal of this series is to get you familiar with engines that run the core ML models, from basic linear and logistic regressions all the way to complex algorithms such as neural networks. <br><br>More details 👇🏼<br><a target=\"_blank\" rel=\"noreferrer nofollow\" href=\"https://ramikrispin.substack.com/p/introduction-to-machine-learning\">https://ramikrispin.substack.com/p/introduction-to-machine-learning</a>","text_length":526,"media":{"root":"/00c/rQcAAMoWZ2EAAAAAeq_pIerVx6E","webpage":{"url":"https://ramikrispin.substack.com/p/introduction-to-machine-learning","type":"photo","title":"Introduction to Machine Learning from Scratch","site_name":"Substack","display_url":"ramikrispin.substack.com/p/introduction-to-machine-learning","description":"Understand the series approach, choose NumPy, MLX, or base R, and follow the path from numerical foundations to machine learning models","thumbs":{"m":{"w":320,"h":160,"hash":"j_Gnch0pshkbAO9Nn8lB3Q&ts=1794905668"},"x":{"w":800,"h":400,"hash":"2-036O91Z7gZE88g9sJGHQ&ts=1794905668"},"y":{"w":1280,"h":640,"hash":"MhpO4avCac2D8CYGqZYwaA&ts=1794905668"},"w":{"w":1600,"h":800,"hash":"t8Ga4MbD5cr8ZXR5lyN7yA&ts=1794905668"},"i":{"bytes":"AUACg|Cb5wx/1g64AHFL8x4zIPwown95x9Go+Qk4Zhk9jTAXLdMyZ68CniTJxhgfcUxSkZyCT25NO80YpAPzRTPNFFAFHefWl3n1oooAAxo3t60UUAG8880UUUAf/9k="}}}}},{"channel_id":1634145994,"post_id":1964,"date":1793712736000,"forwards":"1","views":"49","text":"I feature an open-source project in my newsletter every week, and this week&#039;s pick is DuckPad.<br><br>DuckPad is an open-source SQL workspace powered by DuckDB 2.0. It turns data sources, joins, reusable components, and SQL transformations into visual pipelines that teams can explore, version, compare, and review on their own infrastructure.<br><br>Key features include:<br>✅ Visual pipelines for data sources, joins, and reusable components<br>✅ Guided forms and an integrated SQL editor for transformations<br>✅ Cell-level lineage from result cells to contributing source values<br>✅ Git-backed projects with independent development branches<br>✅ Visual comparison and review of proposed pipeline changes<br>✅ Self-hosted operation with environment-specific connections<br>✅ Team access controls within a shared DuckPad deployment<br><br>More details are available here: <a target=\"_blank\" rel=\"noreferrer nofollow\" href=\"https://github.com/hugues31/duckpad\">https://github.com/hugues31/duckpad</a><br><br>License: MIT<br><br>♻️ Please share if you find it useful<br><br>📌 To receive weekly updates, subscribe to my newsletter: <a target=\"_blank\" rel=\"noreferrer nofollow\" href=\"https://ramikrispin.substack.com/\">https://ramikrispin.substack.com/</a>","text_length":1017,"media":{"root":"/00b/rAcAAMoWZ2EAAAAAgz_tAKfx_ck","photo":{"thumbs":{"m":{"w":320,"h":222,"hash":"w6lGHYGj1gDzLkXnZXDBZQ&ts=1794905668"},"x":{"w":800,"h":556,"hash":"GuGtEZqHLjgPnHGfTZZzng&ts=1794905668"},"y":{"w":1280,"h":889,"hash":"I3PQ9_Ue-kgORKNTmdMdTQ&ts=1794905668"},"i":{"bytes":"AcACg|DWxxSg5FA70gHrzQAuaTrRS0ANI3Y5IPtRTh046UUAHY0mR60Z4NHWgAyvrRkU6igCIM/H7tvzFFS0UXA//9k="}}}}},{"channel_id":1634145994,"post_id":1963,"date":1793548005000,"forwards":"2","views":"53","text":"New tutorial - How to read an LLM model label on Hugging Face 👇🏼<br><br>We usually refer to an open-weight model on Hugging Face by its model family label and parameter count (e.g., Qwen 3.5 - 35B, Gemma 4 31B, etc.), but there&#039;s more to it.<br>One challenge I faced when I started working with local LLMs was understanding Hugging Face&#039;s model labeling system 😵‍💫.<br><br>So, I did a deep dive and created a short tutorial that focuses on that topic - how to parse the model long label.<br><br><a target=\"_blank\" rel=\"noreferrer nofollow\" href=\"https://theaiops.substack.com/p/understanding-local-llm-models\">https://theaiops.substack.com/p/understanding-local-llm-models</a><br><br>This tutorial is part of the MLX series 😎","text_length":584,"media":{"root":"/006/qwcAAMoWZ2EAAAAAYzEd0sxmE0E","webpage":{"url":"https://theaiops.substack.com/p/understanding-local-llm-models","type":"photo","title":"How to Read an LLM Model Label on Hugging Face","site_name":"Substack","display_url":"theaiops.substack.com/p/understanding-local-llm-models","description":"Understand model families, variants, parameter counts, and quantization labels, and learn what to verify beyond the name.","thumbs":{"m":{"w":320,"h":160,"hash":"Q8cY1QNwYIZYNIl3GWPNJQ&ts=1794905668"},"x":{"w":800,"h":400,"hash":"WNRWF1SwfbcQMFzF7aRyrg&ts=1794905668"},"y":{"w":1280,"h":640,"hash":"GtHck-7jfHRqMxQv1X9FDw&ts=1794905668"},"w":{"w":1600,"h":800,"hash":"-SyyRfRFRfH_ucdUm0wq5w&ts=1794905668"},"i":{"bytes":"AUACg|Cxv35wFYHPBoCKT/q4se3/AOqneYgHoBTHlIPAG31zVbkihRtJ8uL2o2puwUixURlb7wAA9c1IkxPLABfXNFguSKcIAAo9l6UUgkQ0UikUWdsdaTcShFFFABn5MdutBYlDRRQSKjHbiiiikM//2Q=="}}}}},{"channel_id":1634145994,"post_id":1962,"date":1793477370000,"forwards":"1","views":"39","text":"My weekly newsletter is out 👇🏼<br><a target=\"_blank\" rel=\"noreferrer nofollow\" href=\"https://ramikrispin.substack.com/p/ai-assisted-python-duckpad-issue\">https://ramikrispin.substack.com/p/ai-assisted-python-duckpad-issue</a><br><br>#ai #dataengineering #datascience","text_length":135,"media":{"root":"/003/qgcAAMoWZ2EAAAAAvwDoUbFJeyY","webpage":{"url":"https://ramikrispin.substack.com/p/ai-assisted-python-duckpad-issue","type":"photo","title":"AI-Assisted Python, DuckPad | Issue 112","site_name":"Substack","display_url":"ramikrispin.substack.com/p/ai-assisted-python-duckpad-issue","description":"Explore Reuven M. Lerner's AI-Assisted Python for Nonprogrammers, DuckPad's visual SQL pipelines and data lineage, agent harnesses, local AI agents on Ubuntu, and prefill-only decision models.","author":"Rami Krispin","thumbs":{"m":{"w":320,"h":160,"hash":"7yCwKMGxq02vOEfGJ8IObQ&ts=1794905668"},"x":{"w":800,"h":400,"hash":"pLdkm32Iv0zluVNpJBSQjw&ts=1794905668"},"y":{"w":1280,"h":640,"hash":"N-ZCTwwep4fVuaJBVoue2A&ts=1794905668"},"w":{"w":1600,"h":800,"hash":"i9KlR0M03iYLH4EFXI82dA&ts=1794905668"},"i":{"bytes":"AUACg|CvvaWDzBwd3b/P+c0uQluyP95zn/69V0Lqqj+HOcVId7ncw6AUXZW+4keUO5eCKnWAzMpYnHXFQgldy7AT6GhjIWU9OMEdqSuPQVkjDsyn5Qc9e2aKgGRGVx1op3sLluK5xGuKfklTz1HNFFDBCHhvw/pTv4hz0oopDQxuEc0UUUDP/9k="}}}}},{"channel_id":1634145994,"post_id":1961,"date":1793366570000,"views":"56","text":"It is a practical resource for data engineers, ML engineers, and MLOps practitioners who want to build and maintain reliable data infrastructure for multimodal AI applications.<br><br>More details are available here: <a target=\"_blank\" rel=\"noreferrer nofollow\" href=\"https://learning.oreilly.com/library/view/data-engineering-for/9781098190774/\">https://learning.oreilly.com/library/view/data-engineering-for/9781098190774/</a><br><br>♻️ Please share if you find it useful<br>📌 Subscribe to my newsletter: <a target=\"_blank\" rel=\"noreferrer nofollow\" href=\"https://ramikrispin.substack.com/\">https://ramikrispin.substack.com/</a>","text_length":392,"media":{"root":"/002/qQcAAMoWZ2EAAAAAhS9KIzvsUO4","webpage":{"url":"https://learning.oreilly.com/library/view/data-engineering-for/9781098190774","type":"photo","title":"Data Engineering for Multimodal AI","site_name":"O’Reilly Online Learning","display_url":"oreilly.com/library/view/data-engineering-for/9781098190774","description":"AI is only as capable as the context it’s given, and context is a data engineering problem before it’s a model problem. As multimodal AI systems and applications become increasingly... - Selection from Data Engineering for Multimodal AI [Book]","author":"Vasundra Srinivasan","thumbs":{"m":{"w":320,"h":168,"hash":"x-AWpnmqSlCRa2x5XTbGCg&ts=1794905668"},"x":{"w":800,"h":420,"hash":"3nXwUUSGnEODUcGEtMcZ0Q&ts=1794905668"},"y":{"w":1200,"h":630,"hash":"xw_c85DeCIGKNVancTMEQw&ts=1794905668"},"i":{"bytes":"AVACg|DZpksixrubPXAxT6jmGVHy55oAhe9jRCeC/wDdzzT7SZp4t7Lt5wPeofIVht8oY9P8mrEKeXGFxgCkPSxLRRRTEFFFFAATxQOlFFABRRRQB//Z"}}}}},{"channel_id":1634145994,"post_id":1960,"date":1793366556000,"views":"54","text":"I feature a data science book every week in my newsletter, and this week&#039;s pick focuses on Data Engineering for Multimodal AI by Vasundra Srinivasan.<br><br>The book presents multimodal AI infrastructure as a data engineering problem. It moves from architecture, ingestion, and feature engineering through retrieval, streaming, MLOps, serving, governance, and security, with practical playbooks that readers can extend across cloud-native and on-premises environments.<br><br>Topics include:<br>✅ Semantic layers, data contracts, and multimodal storage patterns<br>✅ Validation and transformation across data modalities<br>✅ Feature extraction, fusion, and multimodal feature stores<br>✅ Vector search, knowledge graphs, and hybrid retrieval<br>✅ Event alignment, adaptive processing, and edge computing<br>✅ Versioning, testing, monitoring, and automated retraining<br>✅ Distributed processing, inference deployment, and cost management<br>✅ Data lineage, privacy, and responsible AI practices","text_length":958,"media":{"root":"/001/qAcAAMoWZ2EAAAAAVk9ou28Qy0s","photo":{"thumbs":{"m":{"w":244,"h":320,"hash":"2i4UabMd5dB7p1mIck4XwQ&ts=1794905668"},"x":{"w":609,"h":800,"hash":"04ptPqdsqDcJYTUqDjTh2g&ts=1794905668"},"y":{"w":975,"h":1280,"hash":"yyJB1OM7SgbETWJlBE9Cow&ts=1794905668"},"i":{"bytes":"AoAB8|DYwc53H6UtUp5pjc+VDwF+8cZpn2iOBSQS8h6nH86Vx2L+RnGRn0paypLpmlWSJxuxggitC3mE8QbGD0I9DRcLDJFhMxZshwMZHeotkUeRGQF6nI4qwyZcn5sHsDxTfL93/wC+qYiFoYHb5mIPsDUtsiRgqrls88jmnCLd1dx+NSIgTOCTn1oAdRRRQAmT6GgEnsRRRQB//9k="}}}}},{"channel_id":1634145994,"post_id":1959,"date":1793247961000,"views":"17","text":"Next-Concept Modeling (NCM), a new AI architecture explained:<br><a target=\"_blank\" rel=\"noreferrer nofollow\" href=\"https://www.youtube.com/watch?v=_87vrjxx1EE\">https://www.youtube.com/watch?v=_87vrjxx1EE</a>","text_length":105,"media":{"root":"/013/pwcAAMoWZ2EAAAAAJC97GOG_yXw","webpage":{"url":"https://youtube.com/watch?v=_87vrjxx1EE","type":"video","title":"NCM: The End of LLMs? Not So Fast (Local Tests)","site_name":"YouTube","display_url":"youtube.com/watch?v=_87vrjxx1EE","description":"Next-Concept Modeling (NCM) from Stanford’s Hazy Research writes an answer as a rough sketch first and the exact words last, in a handful of passes instead of one pass per token. I ran the released models on a DGX Spark and timed NCM against a same-size word-by-word LLM: it’s up to 4x faster for one request and 4–8x slower at 64 requests. I also trained my own diffusion model, which tied NCM’s 86% on hard Sudoku once it filled the most confident cell first.\n\nPaper: https://arxiv.org/abs/2610.14162\nCode: https://github.com/HazyResearch/ncm\nCheckpoints: https://huggingface.co/hazyresearch/ncm-checkpoints\nHazy Research blog: https://hazyresearch.stanford.edu/blog/2026-10-12-ncm\nSudoku setup (S-FLM): https://github.com/jdeschena/s-flm\nMost-confident-first decoding (Kim et al. 2025): https://arxiv.org/abs/2502.06768\nDiffusion Gemma Locally: https://youtu.be/I2-_3ieJelE\nDiffusion LLMs Are Here!: https://youtu.be/0B9EMddwlOQ\n\nOptional disclosure: Diffusion Gemma and the Gemini logo appear briefly. If you want the line “I’m joining Google; this is based on public information and my views are my own”, confirm the wording with your recruiter.\n\nChapters (timed from the final edit)\n0:00 What NCM (Next-Concept Modeling) is\n0:55 Two questions the paper never tested\n1:57 Why LLMs write one token at a time\n3:07 Diffusion models: the obvious fix\n4:24 How NCM works: coarse-to-fine levels\n5:35 What a “concept” is\n6:04 The generator: one level per pass\n7:06 Paper results: 31x faster, 86% on Sudoku\n7:49 What most coverage missed\n8:58 Test 1: watching NCM solve Sudoku\n10:01 Test 2: NCM vs a same-size LLM\n10:38 64 requests: the result flips\n12:11 Test 3: my own diffusion Sudoku model\n12:40 Most-confident-first ties NCM\n13:48 NCM limitations\n14:52 Bottom line","thumbs":{"m":{"w":320,"h":180,"hash":"XEJii1t7aAtZVr7zLs7waw&ts=1794905668"},"x":{"w":800,"h":450,"hash":"-ThSrqWsdmC5kzx9jEl3bA&ts=1794905668"},"y":{"w":1280,"h":720,"hash":"iKUgTglvkfQSA31M27DehA&ts=1794905668"},"i":{"bytes":"AXACg|BqqiJvcgKKVbmIuFjjZvU46VG8yjarpuRSGcY6/wCc0DUHGfs8JCgdOoHfpWMYJq7N5zd7IsLhlldiuxDww5yKbugxzJjrxtPpmpIyj2SRsVVZAFJB6dTVRkiEa7c4IY4HbPH8qFGPUE5vYkZFkjZ42BC4yCCCM9KKkUokRiijAQg7jnkkd/5UUmrbFxbtqMkiby1QH5WOGOe1WDhIysfGBxRRUC6kVkrJFtfBUdMVIzgcDNFFN7hFaELsSCAxxRRRVJDbP//Z"}},"embed":{"url":"https://www.youtube.com/embed/_87vrjxx1EE","type":"iframe","w":1280,"h":720}}}},{"channel_id":1634145994,"post_id":1958,"date":1793247749000,"forwards":"2","views":"23","text":"AI agent memory explained by the Google Claude team 👇🏼<br><a target=\"_blank\" rel=\"noreferrer nofollow\" href=\"https://www.youtube.com/watch?v=8C4bh59FtsI\">https://www.youtube.com/watch?v=8C4bh59FtsI</a>","text_length":100,"media":{"root":"/012/pgcAAMoWZ2EAAAAABjR8pgXoYHI","webpage":{"url":"https://youtube.com/watch?v=8C4bh59FtsI","type":"video","title":"AI Agent Memory Architecture: Structured SQL, Semantic Vector Search & Memory Bank","site_name":"YouTube","display_url":"youtube.com/watch?v=8C4bh59FtsI","description":"Code for this episode → https://g.dev/ai/fashion-mind\n\nIn this episode, you'll learn the critical difference between Structured Facts and Semantic Understanding, why they should never live in the same database, what to do with raw audio/video perception, and how to read and write memory asynchronously without blocking latency-sensitive voice streams.\n\nWatch along and learn:\n* Short-Term (RAM) vs. Long-Term Agent Memory\n* Facts (Exact Lookup) vs. Understanding (Semantic Recall)\n* Why dumping exact facts into a vector store breaks your agent\n* Where to store Structured Memory (PostgreSQL, MySQL, Google Cloud SQL) vs. Semantic Memory (pgvector, AlloyDB, Vertex AI Agent Engine Memory Bank)\n* How to handle raw multimodal perception (camera frames & audio waveforms)\n* he Two-Track Read/Write Loop: Synchronous tool calls for facts vs. asynchronous off-the-hot-path extraction for understanding\n\nChapters:\n00:00 - Intro: What should a multimodal agent remember?\n00:25 - Demo: Personal styling agent with long-term memory\n00:53 - Short-term (RAM) vs. Long-term memory\n01:28 - The 2 types of memory: Facts vs. understanding\n02:29 - Where memory lives: SQL databases vs. vector stores\n03:40 - Vector store options & managed memory bank\n04:30 - Common mistake: Putting everything in a vector store\n05:24 - What about raw perception? (Audio & video streams)\n06:40 - How agents read memory at session start\n07:38 - Saving memory: Fast sync writes vs. async extraction\n08:44 - Keeping memory extraction off the live hot path\n10:42 - Architecture recap & next episode: Real time vision\n\n💬 Question of the day:\nWhat is the first thing you would want your AI agent to remember about you? Drop your answer in the comments below!\n\n🔔 Subscribe and stay tuned for the next episode, where we give the agent eyes with real-time vision!\n\nResources:\nGemini Live API docs → https://g.dev/cloud/gemini-live\nGoogle Cloud SQL docs → https://g.dev/cloud/sql-docs\nAlloyDB docs → https://g.dev/cloud/alloydb-docs\nVertex AI Agent Engine Memory Bank → https://g.dev/ai/scale-your-agents\n\nFollow Annie online:\nLinkedIn → https://goo.gle/annie-linkedin\nX → https://goo.gle/annie-x\n\nWatch more The Omni App → https://g.dev/cloud/modern-ai-app\n🔔 Subscribe to Google Cloud Tech → https://goo.gle/GoogleCloudTech\n\n#AIAgents #MultimodalAI #SoftwareArchitecture #VectorDatabase #GoogleCloud #LLM #MachineLearning #AgenticAI #CloudSQL #VertexAI\n\nSpeaker:  Annie Wang\nProducts Mentioned: Google Cloud SQL, AlloyDB, Vertex AI Agent Engine Memory Bank, Gemini Live API","thumbs":{"m":{"w":320,"h":180,"hash":"IYCjqV7TcbRL2HDVjwQ6Eg&ts=1794905668"},"x":{"w":800,"h":450,"hash":"v2RAde9dd5b0iz0fmy_Cuw&ts=1794905668"},"y":{"w":1280,"h":720,"hash":"891bZeNVBJES872ZvPNxYw&ts=1794905668"},"i":{"bytes":"AXACg|DVaTa2Npx6ihZCxA2MPc1HJCjSkmQqx5wDSmEAj94/HXnrTESu4RSzdBWbcTzSPIYZCoQjaOx+tXJR/opUMW5AyfrWYgf7ZLH/AA4GfrRpa41uaEF6jwq0nyvjkYoqoyFVzmitIwi1czcmmaDQ7pg+7GO2KXbJ5pO4bfSiismy7CyA+WckH8KpRqvmSOOrHB/KiikykRTP+8xRRRW9PYiW5//Z"}},"embed":{"url":"https://www.youtube.com/embed/8C4bh59FtsI","type":"iframe","w":1280,"h":720}}}},{"channel_id":1634145994,"post_id":1957,"date":1793026099000,"forwards":"1","views":"53","text":"What are Docker Sandboxes? New tutorial 🚀<br><br>I published the first tutorial in my Docker Sandboxes series in the AIOps newsletter, exploring how to give AI coding agents a controlled environment.<br>An agent that edits files, installs packages, runs tests, and builds containers needs broad access. Choosing where those operations run helps limit what a mistake or malicious dependency can reach.<br><br>This first tutorial covers:<br>✅ Why AI agents need isolation<br>✅ MicroVMs and private Docker daemons<br>✅ Workspace and network boundaries<br>✅ Host-side credential handling<br>✅ Use cases and trade-offs<br>✅ Why reviewing changes still matters<br><br><a target=\"_blank\" rel=\"noreferrer nofollow\" href=\"https://theaiops.substack.com/p/what-is-docker-sandboxes\">https://theaiops.substack.com/p/what-is-docker-sandboxes</a>","text_length":680,"media":{"root":"/011/pQcAAMoWZ2EAAAAAc_VhVqkVMKU","webpage":{"url":"https://theaiops.substack.com/p/what-is-docker-sandboxes","type":"photo","title":"What is Docker Sandboxes?","site_name":"Substack","display_url":"theaiops.substack.com/p/what-is-docker-sandboxes","description":"Learn how Docker Sandboxes use microVM isolation, controlled workspaces, private Docker daemons, and network policies to run AI coding agents more safely.","thumbs":{"m":{"w":320,"h":160,"hash":"6CXpwwpqL8IQzkv78qMo6A&ts=1794905668"},"x":{"w":800,"h":400,"hash":"KDEieOjDiMWrk7ro4BtAFg&ts=1794905668"},"y":{"w":1280,"h":640,"hash":"B90GcN30xRjI76H5KhZlGQ&ts=1794905668"},"w":{"w":1600,"h":800,"hash":"I1qEWusFWGf2rR3lKJKWkw&ts=1794905668"},"i":{"bytes":"AUACg|CyVVzuYH6hjTWVFGUUkr6uRik3Z/5aH8hTmaMgg9D15piI1AYDamc+knWpEhVlPmIVPpvNJE8axgJ0xx1pZG3qQpwaYhcJEwOTk8DLGik3hdgyffAopDKYznOT0pSxPB70UUhiA4GBwPal3n1oooELvPNFFFAz/9k="}}}}},{"channel_id":1634145994,"post_id":1956,"date":1792993011000,"forwards":"1","views":"39","text":"OpenCode 2.0 is out, and it looks super awesome! <br><br><a target=\"_blank\" rel=\"noreferrer nofollow\" href=\"https://www.youtube.com/watch?v=OyuId4YQiIs\">https://www.youtube.com/watch?v=OyuId4YQiIs</a>","text_length":94,"media":{"root":"/00e/pAcAAMoWZ2EAAAAA7MNun_1I8lw","photo":{"thumbs":{"m":{"w":320,"h":180,"hash":"r-_dadvbRMdH2ML6sn3PaA&ts=1794905668"},"x":{"w":800,"h":450,"hash":"EQW-IU8xZP-h09TJyiXZSQ&ts=1794905668"},"y":{"w":1280,"h":720,"hash":"XqLQbJCWh0YpvhgfC37cpw&ts=1794905668"},"i":{"bytes":"AXACg|C/NN5cgXzo0J6BjzSTSNFjdNGmem41S1LDXJTyGwQN0irkn6U2dt8qymCR41Qx4I5z60rjsaJZ1tpXdw3ykgj6Vm6bIEQDIGasgNDojK/DbCPzqjYudigHGDnpQ9hx3DVVXzlkU5Ljn8KKTU2zIinnaDziihCklc2nt1dy25lJ9DSmLnlvzFFFDQIxb28eR3hQ4j6fXFQQSCP7y7l+uKKKq2gr6jriYzvuIwo4A9KKKKaQmz//2Q=="}}}}},{"channel_id":1634145994,"post_id":1955,"date":1792973408000,"forwards":"1","views":"31","text":"Multimodal RAG for Images 🚀<br><br>This tutorial from Alison Presmanes Hill provides an example of multimodal RAG for images by building a photo-search app in six ways with Pixeltable.<br><br>📽️: <a target=\"_blank\" rel=\"noreferrer nofollow\" href=\"https://www.youtube.com/watch?v=92lsL395T6s\">https://www.youtube.com/watch?v=92lsL395T6s</a>","text_length":229,"media":{"root":"/00d/owcAAMoWZ2EAAAAA6_Fg6skwlLw","photo":{"thumbs":{"m":{"w":320,"h":180,"hash":"D2lNjC_ibl9osqtv8m9_gg&ts=1794905668"},"x":{"w":800,"h":450,"hash":"jBvaSzynZn738q1BtTlcbQ&ts=1794905668"},"y":{"w":1280,"h":720,"hash":"pQ_TyKBtsv9IKRCg6qS_tg&ts=1794905668"},"i":{"bytes":"AXACg|DNVAQpKvjuQKfshzjE3/fIpYt3lrgMevRqBvwWIfHPcUgK5GOtOSJ5M7FJx6VNJHwWZWJA65FT6fIEJQnG4etA0UCCDg9aKkm5bd60UAWvsCj+JqPsC4++1FFZczLsg+wLnG9qebMbcbjRRRzMLIb9hB6uwooopczCyP/Z"}}}}}]