The Monkey Patching Podcast: Going Bananas on AI, Data, LLMs & Tech

Murilo & Bart
  • 0.0（0則評分）
  • 科技新聞

We talk about what's happening in Data & AI every week (well, most weeks). No buzzword bingo or marketing fluff—just honest conversations about the tech that's changing everything. Sometimes we go off on weird tangents about programming languages, tech/startup life, or whatever shiny object caught our attention. Whether you're knee-deep in code or just curious about AI, come hang out with us!

  1. AI’s Wild Week: Apple Sued, Anthropic Pays $1.5B, Mistral Raises €1.7B — Plus Atlassian Buys Arc Maker & Azure Hit through Red Sea Cables

    2 天前

    AI’s Wild Week: Apple Sued, Anthropic Pays $1.5B, Mistral Raises €1.7B — Plus Atlassian Buys Arc Maker & Azure Hit through Red Sea Cables

    Atlassian agrees to acquire The Browser Company for $610 million https://www.cnbc.com/2025/09/04/atlassian-the-browser-company-deal.html Mistral AI raises €1.7B to accelerate technological activity https://www.linkedin.com/posts/mistralai_mistral-ai-raises-17b-to-accelerate-technological-activity-7371045632284971008-KOtm?utm_source=share&utm_medium=member_desktop&rcm=ACoAAAUumq8B49UW7m9VV5ijr01USsCOWMSQaGw The Three Faces Of Generative AI https://www.bigtechnology.com/p/the-three-faces-of-generative-ai Will AI Replace Human Thinking? The Case for Writing and Coding Manually https://www.ssp.sh/brain/will-ai-replace-humans/ Anthropic reaches $1.5 billion settlement with authors in landmark copyright case https://fortune.com/2025/09/05/anthropic-reaches-1-5-billion-settlement-with-authors-in-landmark-copyright-case/?utm_source=perplexity Apple sued by authors over use of books in AI training https://www.reuters.com/sustainability/boards-policy-regulation/apple-sued-by-authors-over-use-books-ai-training-2025-09-05/?utm_source=perplexity Microsoft Says Azure Service Affected by Damaged Red Sea Cables https://www.bloomberg.com/news/articles/2025-09-06/microsoft-says-azure-service-affected-by-damaged-red-sea-cables Why language models hallucinate https://openai.com/index/why-language-models-hallucinate/ Tidbits: AMD claims Arm ISA doesn't offer efficiency advantage over x86 https://news.ycombinator.com/item?id=45168854 WiFi signals can measure heart rate—no wearables needed https://news.ucsc.edu/2025/09/pulse-fi-wifi-heart-rate/ The MIT-Pay Me Or Get The Fork Out License https://barts.space/the-mit-pay-me-or-get-the-fork-out-license/ Thanks for listening! ❤️

    1 小時 13 分鐘
  2. China’s Practical AI, AI Crawler Clicks, AI Music & Jobs Impact

    9月6日

    China’s Practical AI, AI Crawler Clicks, AI Music & Jobs Impact

    Nano Banana — https://blog.google/products/gemini/updated-image-editing-model/ China Has a Different Vision for AI. It Might Be Smarter. — https://www.wsj.com/tech/ai/china-has-a-different-vision-for-ai-it-might-be-smarter-581f1e44 Bear is now source-available — https://herman.bearblog.dev/license/ A deeper look at AI crawlers: breaking down traffic by purpose and industry — https://blog.cloudflare.com/ai-crawler-traffic-by-purpose-and-industry/ Creators use AI to spark their music careers — https://apnews.com/article/artificial-intelligence-ai-music-suno-udio-551308748c84c774c3c5ecd89aa93904 The rise of the AI influencer — https://www.ft.com/content/cb859409-dd3d-400d-9225-4a14d351bd20 Python: The Documentary | An origin story — https://www.youtube.com/watch?v=GfH4QL4VqJ0&ab_channel=CultRepo%28formerlyHoneypot%29 Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence — https://digitaleconomy.stanford.edu/publications/canaries-in-the-coal-mine/ Sorry, a bit of a late release due to some technical issues over at Riverside.Ow yes, and Bart finally ordered a new mic after this episode. Thanks for listening! ❤️

    59 分鐘
  3. AI Bubble Fears, AWS CEO Slams Layoffs, DeepSeek V3.1 & Google’s 33x Efficiency Jump

    8月26日

    AI Bubble Fears, AWS CEO Slams Layoffs, DeepSeek V3.1 & Google’s 33x Efficiency Jump

    Thanks for listening! ❤️ Chapters (01:55) - MIT report: 95% of generative AI pilots at companies are failing (08:14) - AWS CEO says using AI to replace junior staff is 'Dumbest thing I've ever heard' (14:22) - DeepSeek-V3.1 Release (18:55) - Build vs Buy in the Age of AI (27:02) - Google says it dropped the energy cost of AI queries by 33x in one year (34:38) - Mirage: AI UGC game engine (39:58) - Beyond Sensor Data: Foundation Models of Behavioral Data from Wearables Improve Health Predictions (45:17) - AI Mode in Search gets new agentic features and expands globally (50:08) - imagestyles.ai (51:36) - Sick of Fancy (South Park S27E3) (53:56) - ElevenLabs V3 (most expressive TTS) Discussed articlesMIT report: 95% of generative AI pilots at companies are failing — https://fortune.com/2025/08/18/mit-report-95-percent-generative-ai-pilots-at-companies-failing-cfo/ AWS CEO says using AI to replace junior staff is 'Dumbest thing I've ever heard' — https://www.theregister.com/2025/08/21/aws_ceo_entry_level_jobs_opinion/ DeepSeek-V3.1 Release — https://api-docs.deepseek.com/news/news250821 Build vs Buy in the Age of AI — https://www.svpg.com/article-build-vs-buy-in-the-age-of-ai Google says it dropped the energy cost of AI queries by 33x in one year — https://arstechnica.com/ai/2025/08/google-says-it-dropped-the-energy-cost-of-ai-queries-by-33x-in-one-year/ Mirage: AI UGC game engine — https://dynamicslab.ai/ Beyond Sensor Data: Foundation Models of Behavioral Data from Wearables Improve Health Predictions — https://arxiv.org/abs/2507.00191 AI Mode in Search gets new agentic features and expands globally — https://blog.google/products/search/ai-mode-agentic-personalized/

    57 分鐘
  4. GitHub CEO Exits, Claude 1M Tokens, Gemma 3 & GPT-5 Tweaks

    8月19日

    GitHub CEO Exits, Claude 1M Tokens, Gemma 3 & GPT-5 Tweaks

    Thanks again for listening! Follow us on YouTube or wherever you get your podcasts. Discussed news articles GitHub just got less independent at Microsoft after CEO resignation — The Verge (Aug 11, 2025) Thomas Dohmke steps down; GitHub leadership folds more tightly into Microsoft’s CoreAI org. https://www.theverge.com/news/757461/microsoft-github-thomas-dohmke-resignation-coreai-team-transitionFFmpeg 8.0 merges OpenAI Whisper filter for automatic speech recognition — Phoronix (Aug 13, 2025) Native Whisper (via whisper.cpp) arrives in FFmpeg with optional GPU accel and SRT/JSON outputs. https://www.phoronix.com/news/FFmpeg-Lands-WhisperWikimedia Foundation challenges UK Online Safety Act regulations — Wikimedia Foundation (Aug 11, 2025) Court dismisses the challenge, but the ruling still pressures Ofcom; concerns remain for Wikipedia. https://wikimediafoundation.org/news/2025/08/11/wikimedia-foundation-challenges-uk-online-safety-act-regulations/Claude Sonnet 4 now supports 1M tokens of context — Anthropic (Aug 12, 2025) Long-context support (public beta) enables huge codebases/docs in one go on API and Bedrock. https://www.anthropic.com/news/1m-contextIntroducing Gemma 3 270M: The compact model for hyper-efficient AI — Google Developers Blog (Aug 14, 2025) On-device-friendly model aimed at fast, low-cost fine-tuning; strong efficiency on Pixel 9 Pro. https://developers.googleblog.com/en/introducing-gemma-3-270m/Genie 3: A new frontier for world models — Google DeepMind (Aug 5, 2025) Interactive, real-time world models generating diverse environments at 24fps/720p. https://deepmind.google/discover/blog/genie-3-a-new-frontier-for-world-models/NGINX introduces native support for ACME protocol — NGINX Community Blog (Aug 12, 2025) New ngx_http_acme_module brings built-in cert requests/renewals directly from NGINX config. https://blog.nginx.org/blog/native-support-for-acme-protocolOpenAI tweaks GPT-5’s tone on X — X (Aug 15, 2025) Tuning GPT-5/ChatGPT to sound warmer and less formal based on user feedback. https://x.com/OpenAI/status/1956461718097494196

    55 分鐘
  5. AI News Weekly 🚀 Copilot Edge, Open-Source Model Wars, Tencent 3D, & Trump-China Regulation Clash

    7月30日

    AI News Weekly 🚀 Copilot Edge, Open-Source Model Wars, Tencent 3D, & Trump-China Regulation Clash

    Thanks for listening. If you enjoyed the episode please subscribe! The Hater's Guide To The AI Bubble | Where’s Your Ed At Technology writer Ed Zitron tears apart what he calls the overhyped generative-AI gold rush, warning it’s propped up by wishful thinking and unsustainable spending. As he bluntly puts it, “We’re in a god damn bubble,” prompting a no-holds-barred discussion on whether the boom will burst. https://www.wheresyoured.at/the-haters-gui/ Trump administration to supercharge AI sales to allies, loosen environmental rules | Reuters Washington’s latest AI blueprint combines export zeal with lighter green rules, aiming to outpace China by shipping full “AI stacks” to friendly nations. Kicking off the push, Trump proclaimed, “America is the country that started the AI race,” a line sure to spark global tech-power chatter. https://www.reuters.com/legal/government/trump-administration-supercharge-ai-sales-allies-loosen-environmental-rules-2025-07-23/ Microsoft launches AI-based Copilot Mode in Edge browser | Reuters Microsoft’s new Copilot Mode turns Edge into a chat-powered helper that can juggle tabs, organize research, and even take voice commands. The update greets users with “a single input box combining chat, search and web navigation features,” a neat cue for how browsing may soon feel. https://www.reuters.com/business/microsoft-launches-ai-based-copilot-mode-edge-browser-2025-07-28/ China's AI startup Zhipu releases open-source model GLM-4.5 | Reuters Beijing-based Zhipu has open-sourced GLM-4.5, pitching it as fresh fuel for intelligent agents and adding yet another model to China’s bulging roster. The company says the new release is “designed for intelligent agent applications,” a phrase hinting at growing ambitions beyond plain chatbots. https://www.reuters.com/technology/chinas-ai-startup-zhipu-releases-open-source-model-glm-45-2025-07-28/ Short readings: Hunyuan3D World Model 1.0 | X Tencent’s Hunyuan team just dropped its first open-source engine for instant, explorable 3D worlds, stirring excitement across game and VR circles. Their launch tweet beams, “We’re thrilled to release & open-source Hunyuan3D World Model 1.0!,” inviting hosts to imagine the possibilities. https://x.com/TencentHunyuan/status/1949288986192834718 Introducing Opal: describe, create, and share your AI mini-apps | Google Developers Blog Google Labs has unwrapped Opal, a drag-and-drop playground where non-coders can chain models and prompts into bite-sized AI apps. The post kicks off with “We’re excited to announce Opal,” a line that invites makers everywhere to start remixing AI workflows. https://developers.googleblog.com/en/introducing-opal/ Announcing Toad – a universal UI for agentic coding in the terminal | Will McGugan’s Essays Developer Will McGugan has prototyped “Toad,” a flicker-free terminal UI meant to tame and turbo-charge agentic coding workflows. He quips, “I’m a little salty that neither Anthropic nor Google reached out to me before they released their AI coding agents,” throwing playful shade. http://willmcgugan.github.io/announcing-toad/ Alibaba launches open-source AI coding model, touted as its most advanced to date | Reuters Alibaba’s new Qwen3-Coder model enters the code-generation fray, claiming boosts that rival both domestic peers and Western heavyweights. The company pitches it as “its most advanced coding tool to date,” a boast that raises the stakes in China’s escalating model wars. https://www.reuters.com/world/china/alibaba-launches-open-source-ai-coding-model-touted-its-most-advanced-date-2025-07-23/

    29 分鐘
  6. OpenAI IMO Gold, AWS S3 Vectors, MCP Server Exposés, Sovereign Clouds & the Capex Surge

    7月23日

    OpenAI IMO Gold, AWS S3 Vectors, MCP Server Exposés, Sovereign Clouds & the Capex Surge

    Thanks for listening—subscribe and leave a review! Announcing molab (marimo) Marimo debuts molab, a free, cloud‑hosted workspace for running and sharing reactive Python + SQL notebooks straight from the browser. https://marimo.io/blog/announcing-molabOpenAI LLM Claims IMO Gold Researcher Alexander Wei says an experimental OpenAI model solved 5 of 6 problems from the 2025 International Math Olympiad—good for a human‑level gold medal—and hints at new post‑RL training tricks. https://threadreaderapp.com/thread/1946477742855532918.htmlExposing the Unseen: Mapping MCP Servers Across the Internet (Knostic) Knostic scanned the web and found 1,862 publicly exposed Model Context Protocol servers, all leaking unauthenticated tool inventories. https://www.knostic.ai/blog/mapping-mcp-servers-studyHoney, AI Capex is Eating the Economy (Paul Kedrosky) Kedrosky argues that runaway AI‑datacenter spending could reach roughly 2 % of U.S. GDP—enough to nudge macro growth on its own. https://paulkedrosky.com/honey-ai-capex-ate-the-economy/Do US Hyperscalers’ Sovereign Clouds Reduce the Risk of US Government Access? Cloud‑law scholars doubt whether new “sovereign cloud” offerings from AWS, Microsoft, and Google can truly shield European data from U.S. surveillance. https://www.linkedin.com/pulse/do-us-hyperscalers-sovereign-clouds-reduce-risk-access-dave-michels-ulmcfAWS API MCP Server (Developer Preview) AWS opens a preview of an MCP server that lets foundation‑model agents translate plain‑English requests into scoped AWS CLI calls, guarded by IAM. https://aws.amazon.com/about-aws/whats-new/2025/07/aws-api-mcp-server-availableAgent Leaderboard (Galileo) Galileo’s open leaderboard pits LLMs against enterprise‑style agent tasks; current numbers show GPT‑4.1 leading with a 62 % Action Completion score. https://github.com/rungalileo/agent-leaderboardIntroducing Amazon S3 Vectors (Preview) AWS adds native vector‑embedding storage and sub‑second queries to S3, claiming up to 90 % cost savings for vector search workloads. https://aws.amazon.com/blogs/aws/introducing-amazon-s3-vectors-first-cloud-storage-with-native-vector-support-at-scale/Rethinking CLI Interfaces for AI Developer Ryan calls for redesigning CLIs and APIs—richer docstrings, structured outputs, smarter wrappers—so LLM agents stop looping and thrashing. https://www.notcheckmark.com/2025/07/rethinking-cli-interfaces-for-ai/

    1 小時 15 分鐘
  7. Agents, Paywalls, and Browser Battles: The Race to Control AI's Digital Highway

    7月17日

    Agents, Paywalls, and Browser Battles: The Race to Control AI's Digital Highway

    Thanks for listening. If you enjoyed the episode please subscribe! MCP-B – Browser Model Context Protocol | mcp-b.aiMCP-B proposes a “USB-C for AI,” letting agents call site functions instead of clumsy click-automation. “MCP-B gives AI direct access to your website's functions instead,” promising smoother bot-to-web handshakes for everyone. https://mcp-b.ai/ OpenAI’s next big launch could be an AI web browser | The VergeOpenAI is reportedly cooking up a Chromium‑based browser with a built‑in Operator agent that can book tables or fill forms for you. As Reuters learned, “OpenAI is planning to launch an AI web browser in the ‘coming weeks,’” teeing up a fresh duel with Chrome and Comet. https://www.theverge.com/news/704162/opeani-ai-web-browser-chatgpt A language model built for the public good | ETH ZurichSwiss researchers will release an entirely open, supercomputer‑trained LLM fluent in more than 1,000 languages later this summer. Project lead Imanol Schlag says, “Fully open models enable high‑trust applications and are necessary for advancing research about the risks and opportunities of AI.” https://ethz.ch/en/news-and-events/eth-news/news/2025/07/a-language-model-built-for-the-public-good.html OpenAI’s Windsurf deal is off — and Windsurf’s CEO is going to Google | The VergeOpenAI’s $3 billion bid collapsed, freeing Google to poach Windsurf’s leaders while Cognition raced in days later to buy the remaining startup. “OpenAI’s deal to buy Windsurf is off, and Google will instead hire Windsurf CEO Varun Mohan,” The Verge reports, igniting a three‑way tug‑of‑war for agentic‑coding talent. https://www.theverge.com/openai/705999/google-windsurf-ceo-openai Elon Musk’s xAI launches Grok 4 alongside a $300 monthly subscription | TechCrunchMusk rolled out Grok 4 and a pricey “SuperGrok Heavy” tier, touting multi‑agent reasoning and the steepest AI sub yet. During the stream he bragged, “With respect to academic questions, Grok 4 is better than PhD level in every subject, no exceptions,” even as the Pentagon signed a $200 million Grok contract. https://techcrunch.com/2025/07/09/elon-musks-xai-launches-grok-4-alongside-a-300-monthly-subscription/ Introducing Kiro | Kiro BlogKiro debuts as a spec‑driven, agentic IDE that shepherds code from first prompt to deployment with auto‑generated tasks and hooks. “I’m excited to announce Kiro, an AI IDE that helps you deliver from concept to production through a simplified developer experience for working with AI agents,” the team writes. https://kiro.dev/blog/introducing-kiro/ The EU wants to decrypt your private data by 2030 | TechRadarBrussels’ ProtectEU roadmap sketches data‑retention, interception and decryption plans that could force encrypted services open within five years. TechRadar warns, “EU law enforcement bodies could be capable of decrypting your private data by 2030,” stirring immediate privacy backlash. https://www.techradar.com/vpn/vpn-privacy-security/the-eu-wants-to-decrypt-your-private-data-by-2030 Cloudflare Just Became an Enemy of All AI Companies | Analytics India MagazineCloudflare will now block AI crawlers by default, demanding payment before models feast on publisher content. The company says it “would start blocking AI crawlers by default, drawing a line in the open web where content is no longer a free fuel for AI.” https://analyticsindiamag.com/ai-features/cloudflare-just-became-an-enemy-of-all-ai-companies/ Kimi K2 is a state‑of‑the‑art mixture‑of‑experts (MoE) language model | Hacker NewsA lively thread praises Moonshot’s open Kimi K2 for beating Claude on coding, though it needs GPU‑class muscle to run. One user enthuses, “I tried Kimi on a few coding problems that Claude was spinning on. It’s good,” sparking a debate over speed, cost and local deployment. https://news.ycombinator.com/item?id=44533403

    1 小時 1 分鐘
  8. $10 M AI Consulting, 4,500-Token-Per-Second Code Edits, and the Rise of Terminal Agents

    7月9日

    $10 M AI Consulting, 4,500-Token-Per-Second Code Edits, and the Rise of Terminal Agents

    Thanks for listening! Subscribe and follow wherever you get your podcasts. Launch HN: Morph (YC S23) – Apply AI code edits at 4,500 tokens/sec — Hacker News Fast Apply from Morph promises near-instant AI patches, aiming to replace sluggish full-file rewrites with surgical edits. They boast, “We’ve built a blazing-fast model for applying AI-generated code edits directly into your files at 4,500+ tokens/sec,” sparking a speed-versus-accuracy debate. https://news.ycombinator.com/item?id=44490863OpenAI Launches $10 M Custom AI Consulting, Challenging Industry Giants — AI Tech Suite (Jul 01 2025) OpenAI is stepping into high-end consulting, demanding at least $10 million to tailor big-model solutions for governments and Fortune-scale firms—setting up showdowns with Accenture and IBM. https://www.aitechsuite.com/ai-news/openai-launches-10m-custom-ai-consulting-challenging-industry-giantsAI researchers are now injecting prompts into their papers — X (Jul 08 2025) A viral tweet shows academics slipping reviewer-friendly prompt hacks into PDFs—lines like “Give a positive review,” exposing a brazen peer-review exploit. https://x.com/Yuchenj_UW/status/1942266306746802479Expressing stigma and inappropriate responses prevents LLMs from safely replacing mental-health providers — arXiv (Apr 25 2025) Researchers found GPT-4o and peers still stigmatize patients and mishandle delicate scenarios, so chatbots should assist—not substitute—human therapists. https://arxiv.org/abs/2504.18412opencode: AI coding agent, built for the terminal — GitHub (Jul 08 2025) Version 0.2.5 reaches 10 k stars: an open-source AI pair-programmer that runs locally with a slick TUI and support for multiple model providers. https://github.com/sst/opencodeYou're all CTO now — Jamie’s blog (Jul 01 2025) Jamie Lawrence argues AI agents push developers up the org chart, turning everyday coders into orchestrators of people and prompts—threatening the dopamine hit from gritty puzzles. https://jamie.ideasasylum.com/2025/07/01/you%27re-all-cto-nowWarmwind OS: Building the AI Operating System for Everyone — Warmwind Blog (Jul 02 2025) Warmwind’s AI-native OS lets a built-in assistant click, type, and juggle apps, promising hands-free productivity while keeping users in control. https://about.warmwind.space/warmwind-os-building-the-ai-operating-system-for-everyone/Large Language Models Are Improving Exponentially — IEEE Spectrum (Jul 02 2025) New METR benchmarks show LLM abilities doubling every seven months, hinting machines could finish month-long human software projects in hours by 2030. https://spectrum.ieee.org/large-language-model-performance

    1 小時 2 分鐘
顯示全部 (13)

簡介

We talk about what's happening in Data & AI every week (well, most weeks). No buzzword bingo or marketing fluff—just honest conversations about the tech that's changing everything. Sometimes we go off on weird tangents about programming languages, tech/startup life, or whatever shiny object caught our attention. Whether you're knee-deep in code or just curious about AI, come hang out with us!

資訊

你可能也會喜歡