AI/ML Link Batch — 2026-06-27 Lines 21-100

This page records ingestion of links 21-100 from links-no-label-aah-ai-gold-jjh-strix-2026-06-27.txt. Lines 1-20 were previously ingested in ai-ml-link-batch-2026-06-27. Line 58 was a duplicate of the Karpathy LLM wiki gist already captured in the first batch and was skipped.

Research-interest summary

Most interesting to a CS researcher:

  1. coding-agents-and-skills — agent scaffolding, skill-vs-doc evals, GEPA/DSPy-style text-parameter optimization, and agent execution environments.
  2. local-ai-hardware-and-inference — ROCm, Strix Halo, vLLM/Triton, llama.cpp, and local coding-agent benchmarks as systems evidence.
  3. open-models-and-gemma-4 — open-model progress plus Gemma 4 multi-token prediction as architecture/runtime co-design.
  4. llm-evaluation-and-belief-management — belief revision, forgetting, and statistics for LLM evals.
  5. world-models-and-video-intelligence and robotics-and-embodied-ai — world models, video intelligence, structured 3D latents, and contact-rich robotics.
  6. ai-for-scientific-discovery — AlphaEvolve, synthetic data for differentiable targets, and math/science discovery sources.
LineLinkRaw fileFetch status
21This New Tech Revolutionizes Biology… | Michael Levinraw/articles/021-this-new-tech-revolutionizes-biology-michael-levin.mdfetched via oEmbed
22Overview of Canada’s National Artificial Intelligence Strategy: AI for Allraw/articles/022-overview-of-canada-s-national-artificial-intelligence-strategy-ai-for-all.mdfetched
23AI with AMD ROCm on Ubuntu: your questions answered | Ubunturaw/articles/023-ai-with-amd-rocm-on-ubuntu-your-questions-answered-ubuntu.mdfetched
24mem0 (@mem0ai) on Xraw/articles/024-mem0-mem0ai-on-x.mdfetched
25Everything looks fine at 4-bitraw/articles/025-everything-looks-fine-at-4-bit.mdfetched via oEmbed
26Figure’s First Full HQ Tour: From the Lab to the Factory Floorraw/articles/026-figure-s-first-full-hq-tour-from-the-lab-to-the-factory-floor.mdfetched via oEmbed
27Moravec’s Paradox and the Robot Olympicsraw/articles/027-moravec-s-paradox-and-the-robot-olympics.mdfetched
28Hugo | Substackraw/articles/028-hugo-substack.mdfetched
29stepfun-ai/Step-3.7-Flash-GGUF · Hugging Faceraw/articles/029-stepfun-ai-step-3-7-flash-gguf-hugging-face.mdfetched
30Paper page - When Should Models Change Their Minds? Contextual Belief Management in Large Language Modelsraw/articles/030-paper-page-when-should-models-change-their-minds-contextual-belief-managemen.mdfetched
31How far behind are open models? — LessWrongraw/articles/031-how-far-behind-are-open-models-lesswrong.mdfetched
32GitHub - Luthiraa/TALOS-V2: hardware implementation of transformers running microgpt at 50k+ tkpsraw/articles/032-github-luthiraa-talos-v2-hardware-implementation-of-transformers-running-mic.mdfetched
33DiffusionBlocks: Block-wise Neural Network Training via Diffusion Interpretationraw/articles/033-diffusionblocks-block-wise-neural-network-training-via-diffusion-interpretat.mdfetched
34GitHub - roboflow/rf-detr: RF-DETR is a real-time object detection and segmentation model architecture developed by Roboflow, SOTA on COCO, designed for fine-tuning. (ICLR 2026)raw/articles/034-github-roboflow-rf-detr-rf-detr-is-a-real-time-object-detection-and-segmenta.mdfetched
35Local Coding Agents on Strix Halo and R9700: Pi, Opencode, and SWE-bench Mini Benchmarksraw/articles/035-local-coding-agents-on-strix-halo-and-r9700-pi-opencode-and-swe-bench-mini-b.mdfetched via oEmbed
36Components of A Coding Agentraw/articles/036-components-of-a-coding-agent.mdfetched
37But what exactly are world models?raw/articles/037-but-what-exactly-are-world-models.mdfetched via oEmbed
38Devansh in Artificial Intelligence Made Simpleraw/articles/038-devansh-in-artificial-intelligence-made-simple.mdfetched
39arxiv.org/pdf/2605.19407raw/articles/039-arxiv-org-pdf-2605-19407.mdfetched
40cdn.openai.com/pdf/74c24085-19b0-4534-9c90-465b8e29ad73/unit-distance-remarks.pdfraw/articles/040-cdn-openai-com-pdf-74c24085-19b0-4534-9c90-465b8e29ad73-unit-distance-remark.mdfetched
41AMD Powers Next-Generation Agent Computers with New Ryzen AI Halo Developer Platform and Ryzen AI Max PRO 400 Series Processorsraw/articles/041-amd-powers-next-generation-agent-computers-with-new-ryzen-ai-halo-developer.mdfetched
42Lisan al Gaib (@scaling01) on Xraw/articles/042-lisan-al-gaib-scaling01-on-x.mdfetched
43Using LLM in the shebang line of a scriptraw/articles/043-using-llm-in-the-shebang-line-of-a-script.mdfetched
44Principles and Practice of Deep Representation Learningraw/articles/044-principles-and-practice-of-deep-representation-learning.mdfetched
45We Ran 250 AI Agent Evals to Find Out if Skills Beat Docs. The Answer Is More Complicated Than We Expectedraw/articles/045-we-ran-250-ai-agent-evals-to-find-out-if-skills-beat-docs-the-answer-is-more.mdfetched
46Quantized Neural Networks: The Only Guide You Needraw/articles/046-quantized-neural-networks-the-only-guide-you-need.mdfetched
47A Visual Guide to Gemma 4raw/articles/047-a-visual-guide-to-gemma-4.mdfetched
48Accelerating Gemma 4: faster inference with multi-token prediction draftersraw/articles/048-accelerating-gemma-4-faster-inference-with-multi-token-prediction-drafters.mdfetched
49AlphaEvolve: A coding agent for scientific and algorithmic discoveryraw/articles/049-alphaevolve-a-coding-agent-for-scientific-and-algorithmic-discovery.mdfetched
50Reddit - Please wait for verificationraw/articles/050-reddit-please-wait-for-verification.mdfetched
51Reddit - Please wait for verificationraw/articles/051-reddit-please-wait-for-verification.mdfetched
52GitHub - microsoft/TRELLIS.2: Native and Compact Structured Latents for 3D Generationraw/articles/052-github-microsoft-trellis-2-native-and-compact-structured-latents-for-3d-gene.mdfetched
53Efficient Video Intelligence in 2026raw/articles/053-efficient-video-intelligence-in-2026.mdfetched
54In the AI Era, Shopify Is Investing in Junior Engineers—Not Cutting Them - CoderPadraw/articles/054-in-the-ai-era-shopify-is-investing-in-junior-engineers-not-cutting-them-code.mdfetched
55Neural Garbage Collection: Learning to Forget while Learning to Reasonraw/articles/055-neural-garbage-collection-learning-to-forget-while-learning-to-reason.mdfetched
56Vision Banana | Google DeepMindraw/articles/056-vision-banana-google-deepmind.mdfetched
57Ask The Experts #2 – Lemonade: Fast GenAI on Ryzen AI and Radeonraw/articles/057-ask-the-experts-2-lemonade-fast-genai-on-ryzen-ai-and-radeon.mdfetched via oEmbed
58https://gist.github.com/karpathy/442a6bf555914893e9891c11519de94falready ingested: 01-llm-wiki-github.mdskipped duplicate already in wiki
59lmstudio-community/Qwen3-Coder-Next-GGUF · Hugging Faceraw/articles/059-lmstudio-community-qwen3-coder-next-gguf-hugging-face.mdfetched
60How we OCR’ed 30,000 papers using Codex, open OCR models and Jobsraw/articles/060-how-we-ocr-ed-30-000-papers-using-codex-open-ocr-models-and-jobs.mdfetched
61Neural Computersraw/articles/061-neural-computers.mdfetched
62Synthetic Data for any Differentiable Targetraw/articles/062-synthetic-data-for-any-differentiable-target.mdfetched
63Maziyar PANAHI (@MaziyarPanahi) on Xraw/articles/063-maziyar-panahi-maziyarpanahi-on-x.mdfetched
64Manim Communityraw/articles/064-manim-community.mdfetched
65Consulting the Oracle: Claude on the Future of Dataraw/articles/065-consulting-the-oracle-claude-on-the-future-of-data.mdfetched
66How to Run Local LLMs with Claude Code | Unsloth Documentationraw/articles/066-how-to-run-local-llms-with-claude-code-unsloth-documentation.mdfetched
67Day 0 Support for Gemma 4 on AMD Processors and GPUsraw/articles/067-day-0-support-for-gemma-4-on-amd-processors-and-gpus.mdfetched
68Gemma 4: Byte for byte, the most capable open modelsraw/articles/068-gemma-4-byte-for-byte-the-most-capable-open-models.mdfetched
69Bonsai - a prism-ml Collectionraw/articles/069-bonsai-a-prism-ml-collection.mdfetched
70Release 4.0.0 · huggingface/transformers.jsraw/articles/070-release-4-0-0-huggingface-transformers-js.mdfetched
71GitHub - google-research/timesfm: TimesFM (Time Series Foundation Model) is a pretrained time-series foundation model developed by Google Research for time-series forecasting.raw/articles/071-github-google-research-timesfm-timesfm-time-series-foundation-model-is-a-pre.mdfetched
72Reddit - Please wait for verificationraw/articles/072-reddit-please-wait-for-verification.mdfetched
73DexDrummer: In-Hand, Contact-Rich, and Long-Horizon Dexterous Robot Drummingraw/articles/073-dexdrummer-in-hand-contact-rich-and-long-horizon-dexterous-robot-drumming.mdfetched
74www.nature.com/articles/s41586-026-10265-5.pdfraw/articles/074-www-nature-com-articles-s41586-026-10265-5-pdf.mdfetched
75TurboQuant: Redefining AI efficiency with extreme compressionraw/articles/075-turboquant-redefining-ai-efficiency-with-extreme-compression.mdfetched
76LeWorldModel: Stable End-to-End Joint-Embedding Predictive Architecture from Pixelsraw/articles/076-leworldmodel-stable-end-to-end-joint-embedding-predictive-architecture-from.mdfetched
77arxiv.org/pdf/2603.19312v1raw/articles/077-arxiv-org-pdf-2603-19312v1.mdfetched
78Claude Cowork: Claude Code power for knowledge work | Claude by Anthropicraw/articles/078-claude-cowork-claude-code-power-for-knowledge-work-claude-by-anthropic.mdfetched
79AI is redrawing the database market | ClickHouseraw/articles/079-ai-is-redrawing-the-database-market-clickhouse.mdfetched
80A Visual Guide to Attention Variants in Modern LLMsraw/articles/080-a-visual-guide-to-attention-variants-in-modern-llms.mdfetched
81Skill Issue: Andrej Karpathy on Code Agents, AutoResearch, and the Loopy Era of AIraw/articles/081-skill-issue-andrej-karpathy-on-code-agents-autoresearch-and-the-loopy-era-of.mdfetched via oEmbed
82Consciousness in humans and other things with Anil K Seth | The Royal Societyraw/articles/082-consciousness-in-humans-and-other-things-with-anil-k-seth-the-royal-society.mdfetched via oEmbed
83Introducing Unsloth Studio | Unsloth Documentationraw/articles/083-introducing-unsloth-studio-unsloth-documentation.mdfetched
84SQLite FTS5 Extensionraw/articles/084-sqlite-fts5-extension.mdfetched
85Run OpenClaw Locally On AMD Ryzen AI Max+ And Radeon GPUsraw/articles/085-run-openclaw-locally-on-amd-ryzen-ai-max-and-radeon-gpus.mdfetched
86GEPA: optimize_anything: A Universal API for Optimizing any Text Parameterraw/articles/086-gepa-optimize-anything-a-universal-api-for-optimizing-any-text-parameter.mdfetched
87Hermes Agent Documentation | Hermes Agentraw/articles/087-hermes-agent-documentation-hermes-agent.mdfetched
88Anthropic Coursesraw/articles/088-anthropic-courses.mdfetched
89Bryan Catanzaro (@ctnzr) on Xraw/articles/089-bryan-catanzaro-ctnzr-on-x.mdfetched
90Strix Halo GPU Benchmarksraw/articles/090-strix-halo-gpu-benchmarks.mdfetched
91GitHub - microsoft/markitdown: Python tool for converting files and office documents to Markdown.raw/articles/091-github-microsoft-markitdown-python-tool-for-converting-files-and-office-docu.mdfetched
92Applying Statistics to LLM Evaluationsraw/articles/092-applying-statistics-to-llm-evaluations.mdfetched
93How is hardware reshaping LLM design?raw/articles/093-how-is-hardware-reshaping-llm-design.mdfetched via oEmbed
94THE 2028 GLOBAL INTELLIGENCE CRISISraw/articles/094-the-2028-global-intelligence-crisis.mdfetched
95Reasoning Models Generate Societies of Thoughtraw/articles/095-reasoning-models-generate-societies-of-thought.mdfetched
96GitHub - open-webui/open-terminal: A computer you can curl ⚡raw/articles/096-github-open-webui-open-terminal-a-computer-you-can-curl.mdfetched
97vLLM Triton Attention Backend Deep Diveraw/articles/097-vllm-triton-attention-backend-deep-dive.mdfetched
98AI Enabled Software Development and Jevons Paradox - Mike Grouchyraw/articles/098-ai-enabled-software-development-and-jevons-paradox-mike-grouchy.mdfetched
99Complete DSPy Tutorial - Master LLM Prompt Programming in 8 amazing examples!raw/articles/099-complete-dspy-tutorial-master-llm-prompt-programming-in-8-amazing-examples.mdfetched via oEmbed
100llama.cpp with ROCmraw/articles/100-llama-cpp-with-rocm.mdfetched

Notes

Social, Reddit, PDF, and some JavaScript-heavy pages may have only partial public metadata/text captures. The raw files preserve URL, final URL, line number, fetch status, and a text excerpt when available.