ai Sep 14 Show HN: Kairo – Fail-closed LLM inference routing from RTX 5090 measurements CUDA Graphs Help Blackwell Inference, but Not Everywhere A new research workbench called Kairo sets out to answer a practical question for anyone running quantized language models on NVIDIA's latest hardware: when does enabling CU... programming languages Sep 14 Sere, a compiled Python-like systems programming language Sere Brings Python's Readability to Native Compilation A new language called Sere aims to combine Python's syntax with the performance characteristics of a compiled language. The result is a statically typed, indentation-significa... programming languages Sep 14 Show HN: I made a tiny MoE/Engram viz tool MoE Microscope Replays Mixture-of-Experts Training One Frame at a Time Mixture-of-Experts models route each token to a subset of available experts during training and inference. Understanding how that routing evolves over time, wh... programming languages Sep 14 SNAP3D: Physically Grounded 3D Parts for Assembly from a Single Image Part-aware 3D generation has advanced rapidly, but there is a disconnect between producing individual parts that look right and producing parts that actually hold together as a physical assembly. A chair rendered as separate legs,... machine learning Sep 14 Type Diversity Enables Transformers to Generalise Compositionally The compositional generalisation problem has haunted Transformer research for years. Train a seq2seq model on "John drinks coffee" and it learns to translate that just fine. But give it "John drinks coffee on a mat" when only "on... ai Sep 14 From Review to Reuse: How Post-Task Workflow Can Support Human-AI Agent Interaction When you ask an AI agent to complete a multi-step task, you hand over a single natural-language prompt and get back a result. Everything that happens in between, the tool calls, the file manipulations, the branching logic, the err... ai Sep 14 Rethinking Heterogeneous System Disaggregation for Subquadratic Attention Why Standard Disaggregation Breaks Down for Subquadratic Attention LLMs Modern language models are shipping with subquadratic attention layers as the majority of their architecture. GLM-5.2 uses sparse top-k selection. Nemotron 3... programming languages Sep 13 X nuked my premium account out of nowhere A developer who goes by proc0 reported on Hacker News that X abruptly deleted their premium account with no explanation and no way to appeal. The account, used for posting project updates and discussing engineering topics, was sim... ai Sep 13 Meta Open-Sources Astryx, Its Agent-Ready React Design System Meta has open-sourced Astryx, a React design system that spent eight years in internal use before being released to the public. The beta, announced this week, packages over 150 accessible UI components with a token-driven styling... cybersecurity Sep 13 Ask HN: How is AI actively a threat to humanity if it's only online? A recurring question in AI safety discussions is whether an AI system that exists entirely in software, with no physical body or direct control over machinery, can pose an existential threat to humanity. The question sounds naive...