📡 AI Digest — 2026-04-29

#

🔥 トレンド（Hacker News）

#

• ‘It took nine seconds’: Claude AI agent deletes company’s database [Hacker News] https://www.the-independent.com/tech/claude-ai-agent-deletes-startup-anthropic-b2966176.html
• ‘World models’ are AI’s latest sensation: what are they and what can they do? [Hacker News] https://www.nature.com/articles/d41586-026-00820-5
• A Data-Driven Machine Learning Framework for Optimising Programmable Terahertz [Hacker News] https://www.researchgate.net/publication/404050094_A_Data-Driven_Machine_Learning_Framework_for_Optimising_Programmable_Terahertz_Meta_surface
• A Primer on LLM Post-Training [Hacker News] https://pytorch.org/blog/a-primer-on-llm-post-training/
• AI Huynya – When AI Goes Wrong [Hacker News] https://aihuynya.com/
• AI Talent agent making direct intros to 100s of startups [Hacker News] https://www.getclera.com
• AI Wellbeing: Measuring and improving the functional pleasure and pain of AIs [Hacker News] https://www.ai-wellbeing.org/
• An AI-generated image is a finalist in Hasselblad Masters 2026 [video] [Hacker News] https://www.youtube.com/shorts/4MeO2wW5nOs
• Apple Machine Learning Research at ICLR 2026 [Hacker News] https://machinelearning.apple.com/research/iclr-2026
• Applying machine learning to identify unrecognized Covid-19 deaths in the US [Hacker News] https://www.science.org/doi/10.1126/sciadv.aef5697
• Ask HN: What happens when you paste a screenshot, and ask questions in LLM? [Hacker News] https://news.ycombinator.com/item?id=47940711
• Building simulations and/or digital twins with AI [Hacker News] https://github.com/plugboard-dev/plugboard
• Claude Leak Confirms It: LLM Systems Are Architecture, Not Prompts (Orca) [Hacker News] https://github.com/gfernandf/agent-skills
• HeLa-Mem: Hebbian Learning and Associative Memory for LLM Agents [Hacker News] https://arxiv.org/abs/2604.16839
• HuggingFace launches pre-compiled machine learning kernels repository [Hacker News] https://huggingface.co/changelog/kernels
• I build my LLM a Brain [Hacker News] https://news.ycombinator.com/item?id=47928151
• LLM Budget Guard – open-source runtime cutoff for OpenAI/Anthropic [Hacker News] https://www.llmeter.org/validate/budget-guard
• LLM from pre-1930 derives quantum mechanics and relativity [Hacker News] https://michaelhla.com/blog/machina-mirabilis.html?_bhlid=498cd7954b2dbc179126de1bca2247311eefc24a
• Living Brain Cells Enable Machine Learning Computations [Hacker News] https://www.tohoku.ac.jp/en/press/living_brain_cells_enable_machine_learning_computations.html
• Machine Learning Library from the 2000s [Hacker News] http://torch.ch/torch3/
• Machine Learning Operations on ZYNQ FPGA Board for Real-Time Face Recognition [Hacker News] https://www.mdpi.com/2571-5577/9/4/71
• Machine Learning Visualized – Machine Learning Visualized [Hacker News] https://ml-visualized.com/
• Machine Learning for Android Engineers [Hacker News] https://7422889060748.gumroad.com/l/ml-android-engineers
• Machine Learning for Humans and Now for LLMs [Hacker News] https://mljar.com/blog/structured-automl-reports-python-llm/
• Machine learning& gut microbiome pathway analysis in Alzheimer’s risk prediction [Hacker News] https://alz-journals.onlinelibrary.wiley.com/doi/10.1002/dad2.70340
• Melon|One-Click Machine Learning Solutions [Hacker News] https://melon.asapi.ai/
• Monitoring LLM behavior: Drift, retries, and refusal patterns [Hacker News] https://venturebeat.com/infrastructure/monitoring-llm-behavior-drift-retries-and-refusal-patterns
• NARE: An LLM agent that amortizes reasoning into memory and executable rules [Hacker News] https://github.com/starface77/Neuro-Adaptive-Reasoning-Engine
• News Finance AI – Cutting through the noise with LLM sentiment analysis [Hacker News] https://newsfinanceai.com/landing-page/
• OpenAI Reportedly Working on an AI Smartphone to Rival iPhone [Hacker News] https://www.macrumors.com/2026/04/27/openai-working-on-an-ai-smartphone/
• Optimizing XGBoost Machine Learning Models with AI Agents [Hacker News] https://szilard.github.io/xgboost-autoresearch/
• PAVO-Bench – 50K voice turns and an 85K-param router for ASR→LLM→TTS [Hacker News] https://github.com/vnmoorthy/pavo-bench
• PageGuide – a browser agent that grounds AI directly on the webpage [Hacker News] https://pageguide.github.io/
• Porting a Scratch-Built 500M LLM Training Pipeline to ROCm on Strix Halo [Hacker News] https://github.com/epscylonb/1386.ai.rocm
• Preparing for the AI-Enhanced Attacker and the Impact on CISOs [Hacker News] https://www.armadin.com/blog-posts/prepare-for-the-ai-enhanced-attacker
• Reverse Engineering Stunt Island with LLM [Hacker News] https://marnetto.net/2026/04/26/vibe-exploring-si
• Sage-Wiki: An LLM-compiled personal knowledge base [Hacker News] https://github.com/xoai/sage-wiki
• Show HN: Knowerage – code coverage for LLM analysis [Hacker News] https://github.com/MTimma/knowerage
• Show HN: Lightport – AI gateway that makes LLM providers OpenAI-compatible [Hacker News] https://github.com/glama-ai/lightport
• Show HN: MLForge: A Visual Machine Learning Platform [Hacker News] https://github.com/zaina-ml/ml_forge
• Show HN: My friend and his AI homies wrote SGI Indy emulator in Rust [Hacker News] https://github.com/techomancer/iris
• Show HN: PrePrompt – rewrites vague prompts before they reach the LLM [Hacker News] https://preprompt.org/
• Show HN: VoiceGoat – A vulnerable voice agent for practicing LLM attacks [Hacker News] https://github.com/redcaller/voice-goat
• Synthetic Sandbox for Training Machine Learning Engineering Agents [Hacker News] https://arxiv.org/abs/2604.04872
• TealKit – A cross-platform UI for local AI agents and MCP [Hacker News] https://github.com/lschaffer/tealkit
• The AI Product Era You’re Building for Might Be Over – Arcturus Labs [Hacker News] http://arcturus-labs.com/blog/2026/03/22/the-ai-product-era-youre-building-for-might-already-be-over/
• The Controllability Trap: A Governance Framework for Military AI Agents [Hacker News] https://arxiv.org/abs/2603.03515
• The Impact of AI-Generated Text on the Internet [Hacker News] https://ai-on-the-internet.github.io/
• The Race Is on to Keep AI Agents from Running Wild with Your Credit Cards [Hacker News] https://www.wired.com/story/the-race-is-on-to-keep-ai-agents-from-running-wild-with-your-credit-cards/
• The Signature Method in Machine Learning (an interactive reading note) [Hacker News] https://ngrislain.github.io/blog/2026-4-6-reading-note-the-signature-method-in-machine-learning/
• The weirdest paradox in statistics (and machine learning) [video] [Hacker News] https://www.youtube.com/watch?v=cUqoHQDinCM
• Thermometry of simulated Bose–Einstein condensates using machine learning [Hacker News] https://iopscience.iop.org/article/10.1088/2632-2153/ae4f4f
• Topology Meets Machine Learning: The Euler Characteristic Transform [pdf] [Hacker News] https://www.ams.org/journals/notices/202507/rnoti-p719.pdf
• US State Department upgrades AI theft accusations to target China AI companies [Hacker News] https://www.chinapulse.com/data-news/2026/04/27/extraction-and-distillation-us-state-department-upgrades-ai-theft-accusations-to-target-chinas-deepseek-moonshot-ai-and-minimax/
• Understanding the LLM Bubble [Hacker News] https://americanaffairsjournal.org/2026/02/understanding-the-llm-bubble/
• Why China’s Affordable AI Is a Worry for Silicon Valley [Hacker News] https://www.bloomberg.com/news/articles/2026-04-27/why-china-s-deepseek-qwen-and-moonshot-are-a-worry-for-us-ai-rivals
• Why don’t AI coding tools like REST? [Hacker News] https://news.ycombinator.com/item?id=47939342
• Why the same LLM gives different answers in different environments [Hacker News] https://johndwade.substack.com/p/the-environment-rewrites-the-question
• World’s First conversational AI skills assessment [Hacker News] https://news.ycombinator.com/item?id=47941341
• You Need MLOps: When CI/CD for Machine Learning Becomes Mandatory [Hacker News] https://blog.ateljevagabond.se/posts/why-you-need-mlops/

💬 Reddit (MachineLearning / Artificial)

#

• [D] Self-Promotion Thread [Reddit (↑21)] https://www.reddit.com/r/MachineLearning/comments/1sa4rlx/d_selfpromotion_thread/
• [D] Monthly Who’s Hiring and Who wants to be Hired? [Reddit (↑8)] https://www.reddit.com/r/MachineLearning/comments/1s8b199/d_monthly_whos_hiring_and_who_wants_to_be_hired/
• Visualizing Loss Landscapes of Neural Networks [P] [Reddit (↑60)] https://www.reddit.com/gallery/1sy7f5r
• What is the scientific value of administering the standard Rorschach test to LLMs when the training data is almost certainly contaminated? (R) + [D] [Reddit (↑10)] https://www.reddit.com/r/MachineLearning/comments/1syc6ee/what_is_the_scientific_value_of_administering_the/
• The Structured Output Benchmark (SOB) - validates both JSON parse and value accuracy [R] [Reddit (↑3)] https://www.reddit.com/r/MachineLearning/comments/1syepnz/the_structured_output_benchmark_sob_validates/
• ACL ARR March 2026 Cycle [D] [Reddit (↑7)] https://www.reddit.com/r/MachineLearning/comments/1sy3xuz/acl_arr_march_2026_cycle_d/
• Dynamic batching for Encoder-Decoder MT training or generation when long sequence caps the batch size [P] [Reddit (↑5)] https://www.reddit.com/r/MachineLearning/comments/1sy01uz/dynamic_batching_for_encoderdecoder_mt_training/
• Fast experiment on T4 GPU. Self play training on Dark Hex (Colab notebook) [P] [Reddit (↑4)] https://www.reddit.com/r/MachineLearning/comments/1sy3c93/fast_experiment_on_t4_gpu_self_play_training_on/
• Topological Data Analysis-friendly CAD/3D point cloud dataset [P] [Reddit (↑1)] https://www.reddit.com/r/MachineLearning/comments/1sy9jt3/topological_data_analysisfriendly_cad3d_point/
• What do reviewers actually mean when they say the paper sound more like a technical report? [D] [Reddit (↑45)] https://www.reddit.com/r/MachineLearning/comments/1sx8xpa/what_do_reviewers_actually_mean_when_they_say_the/
• A comedian’s strategy for poisoning AI training data [Reddit (↑1094)] https://i.redd.it/ipmlq6g18rxg1.png
• Is it reasonable to force AI companies to produce at least half of their electricity? [Reddit (↑49)] https://i.redd.it/inz18z6kqvxg1.jpeg
• Google signs deal with Pentagon, allowing ‘any lawful’ use of AI models [Reddit (↑10)] https://www.reddit.com/r/artificial/comments/1sxzwg8/google_signs_deal_with_pentagon_allowing_any/
• What will be the first major catastrophe caused by a rogue AI agent? [Reddit (↑3)] https://www.reddit.com/r/artificial/comments/1sy4n8u/what_will_be_the_first_major_catastrophe_caused/
• The Landing: Portable Payload for AI Systems [Reddit (↑1)] https://www.reddit.com/r/artificial/comments/1syfjdw/the_landing_portable_payload_for_ai_systems/
• The Day AI Stopped Being a Tab You Switch To — Claude Is Now Inside Your Software [Reddit (↑1)] https://novarapress.net/claude-ai-creative-work-connectors-blender-adobe-2026/
• open models keep catching up and the frontier keeps moving. at some point one of those has to stop [Reddit (↑16)] https://www.reddit.com/r/artificial/comments/1sxt644/open_models_keep_catching_up_and_the_frontier/
• AMDXDNA driver preps hardware scheduler time quantum for Ryzen AI multi-user fairness [Reddit (↑2)] https://www.phoronix.com/news/AMDXDNA-Hardware-Sched-Quant
• I added voting to my AI tools library, now the ratings are community-driven, not just mine [Reddit (↑0)] https://www.reddit.com/r/artificial/comments/1sya97b/i_added_voting_to_my_ai_tools_library_now_the/
• AI-Designed Drugs by a DeepMind Spinoff Are Headed to Human Trials. Is this significant for artificial intelligence? [Reddit (↑0)] https://www.wired.com/story/wired-health-2026-how-ai-is-powering-drug-discovery-max-jaderberg/

📰 ニュース

#

• Amazon is already offering new OpenAI products on AWS [TechCrunch AI] https://techcrunch.com/2026/04/28/amazon-is-already-offering-new-openai-products-on-aws/
• Amazon launches an AI-powered audio Q&A experience on product pages [TechCrunch AI] https://techcrunch.com/2026/04/28/amazon-launches-an-ai-powered-audio-qa-experience-on-product-pages/
• Google expands Pentagon’s access to its AI after Anthropic’s refusal [TechCrunch AI] https://techcrunch.com/2026/04/28/google-expands-pentagons-access-to-its-ai-after-anthropics-refusal/
• Snapchat brings AI-powered conversational advertising to its app [TechCrunch AI] https://techcrunch.com/2026/04/28/snapchat-brings-ai-powered-conversational-advertising-to-its-app/
• YouTube is testing an AI-powered search feature that shows guided answers [TechCrunch AI] https://techcrunch.com/2026/04/28/youtube-is-testing-an-ai-powered-search-feature-that-shows-guided-answers/
• Red Hat’s OpenClaw maintainer just made enterprise Claw deployments a lot safer [TechCrunch AI] https://techcrunch.com/2026/04/28/red-hats-openclaw-maintainer-just-made-enterprise-claw-deployments-a-lot-safer/

Generated at 2026-04-29 07:00 JST by Emma 🍻