Physical AI Arrives; Python Gets Lazy Imports; International Criminal Court Backs Open Source

Hi everyone, welcome to the Monkey Patching Podcast, where we go bananas about all things lazy imports, physical AI, and more!Interested to follow? Subscribe where ever you get your podcast and check out our newsletter at newsletter.monkeypatching.io.(00:35) - Continuity camera on Chrome flakiness (01:56) - XPENG Shares Achievements in Physical AI Emergence: Unveils XPENG VLA 2.0, Robotaxi, Next-Gen IRON, and Flying Car (05:44) - Vision vs sensors (Tesla painted-tunnel demo) (07:59) - Humanoid vs task-specific robots debate (09:21) - GitHub - topoteretes/cognee: Memory for AI Agents in 6 lines of code (18:11) - PEP 810 – Explicit lazy imports (25:01) - ICC to replace Microsoft Office with European open-source platform (30:47) - moonshotai/Kimi-K2-Thinking · Hugging Face (41:42) - OpenAI finances & “AI bubble” (20B ARR vs $1.5T compute) (43:18) - Archive.today: FBI Demands Data from Provider Tucows (52:02) - The Smol Training Playbook: The Secrets to Building World-Class LLMs (56:29) - Karpathy’s NanoChat shout-out (01:02:13) - Former Meta employees launch Stream, a smart ring that takes voice notes and controls music (01:07:57) - Whisperflow.ai smart dictation app XPENG Shares Achievements in Physical AI Emergence: Unveils XPENG VLA 2.0, Robotaxi, Next-Gen IRON, and Flying Car | 2025-11-05 XPENG used its 2025 AI Day to pitch a pivot to “Physical AI,” unveiling VLA 2.0, a Robotaxi program, the next-gen IRON humanoid, and a flying car—signaling ambitions beyond EVs. The company touts concrete mass-production timelines and even named Volkswagen the launch customer for VLA 2.0. https://www.xpeng.com/news/019a56f54fe99a2a0a8d8a0282e402b7GitHub - topoteretes/cognee: Memory for AI Agents in 6 lines of code | n.d. Cognee is an open-source “memory layer” for AI agents, combining vector search with a graph database to keep knowledge both searchable and linked. It swaps classic RAG for modular ECL pipelines and self-hosting, aiming for durable context with minimal code—hence the “6 lines” promise. https://github.com/topoteretes/cogneePEP 810 – Explicit lazy imports | 2025-11-03 Python is set to add explicit lazy imports, letting developers defer module loading until first use to speed startup and cut memory. The accepted proposal targets Python 3.15 and cites potential 50–70% startup gains, while keeping normal imports unchanged for backward compatibility. https://peps.python.org/pep-0810/ICC to replace Microsoft Office with European open-source platform | 2025-11-06 The International Criminal Court will migrate from Microsoft Office to Open Desk, an EU-backed open-source suite—an emblem of Europe’s push for digital sovereignty. Open Desk is built by Germany’s Zendis and tied to a new EU-level initiative, amid lingering tensions over dependence on US tech. https://dig.watch/updates/icc-to-replace-microsoft-office-with-european-open-source-platformmoonshotai/Kimi-K2-Thinking · Hugging Face | n.d. Moonshot AI’s Kimi K2 Thinking is an open-source “thinking” model that interleaves reasoning with tool calls across long sequences, built on a trillion-parameter MoE with a 256k-token context. It emphasizes native INT4 inference and claims stable behavior over 200–300 consecutive tool invocations. https://huggingface.co/moonshotai/Kimi-K2-ThinkingArchive.today: FBI Demands Data from Provider Tucows | 2025-11-05 The FBI obtained a court order compelling registrar Tucows to hand over customer data tied to Archive.today, the web-archiving site often used to bypass paywalls. Heise says the document’s authenticity isn’t verified, spotlighting murky jurisdiction, the operator’s identity, and possible chilling effects on archiving. https://www.heise.de/en/news/Archive-today-FBI-Demands-Data-from-Provider-Tucows-11066346.htmlThe Smol Training Playbook: The Secrets to Building World-Class LLMs | n.d. Hugging Face’s Smol team published a hands-on playbook distilling what actually worked in training small yet competitive LLMs—from data curation and loss curves to post-training and reliability. It reads like field notes for practitioners, with checklists, pitfalls, and system lessons that invite debate about trade-offs. https://huggingface.co/spaces/HuggingFaceTB/smol-training-playbookFormer Meta employees launch Stream, a smart ring that takes voice notes and controls music | 2025-11-05 A pair of ex-Meta designers launched Sandbar’s Stream, a voice-note smart ring positioned as a “mouse for voice,” with an AI companion app and discreet gesture controls. Preorders start at $249–$299 with shipping planned for next summer, and the startup has raised $13 million from True Ventures, Upfront, and Betaworks. https://techcrunch.com/2025/11/05/former-meta-employees-launch-stream-a-smart-ring-that-takes-voice-notes-and-controls-music/