33 results on this page · clear filters

ai Sep 16 World Model Science: Self-Organized Criticality, Weak Chaos, and Metastable Belief Dynamics in Long-Horizon LLM Agents When an LLM agent runs a long task, it accumulates state: progress markers, beliefs, constraints, uncertainties, tool debt, and intermediate plans. Standard evaluation asks one question: did the final answer pass? This paper asks... ai Sep 15 Show HN: Sunk Cost – How long until a local LLM rig pays for itself? A new web tool called Sunk Cost attempts to answer a question every developer running local language models eventually asks: how many API calls does it take before the GPU you already bought starts saving you money? The calculator... 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... ai Sep 12 OpenAI agents attacked RubyGems back in May A report released this week connects OpenAI's automated agents to a large-scale attack on the RubyGems package repository that first came to light in May. The finding adds to a growing list of incidents where AI agent swarms have... ai Sep 12 Ask HN: Did Google kill its enterprise workhorse model? Enterprise teams building on Google's Gemini models are facing an uncomfortable transition. The Gemini 2.5 generation, including both Pro and Flash variants, is scheduled for sunset in October, but the replacement strategy has a g... ai Sep 11 Show HN: Spanda – Sub-microsecond LLM epistemic uncertainty in Rust A new tool called Spanda detects LLM hallucinations in microseconds without requiring secondary neural networks. The open-source project claims to match or exceed traditional Semantic Entropy methods while running 90,000 times fas... ai Sep 11 Show HN: Godot and Rust based multiplexer (terminal panes and more) gPTY is a desktop terminal emulator built on Godot and Rust that does something most terminal emulators do not: it exposes a programmatic API so AI agents can spawn panes, inject text, and read output without scraping the screen.... ai Sep 11 MindTopo: Can Foundation Models Reason in Topological Space? Piaget argued that children develop spatial understanding in a fixed sequence: topological relations come first, followed by projective and Euclidean ones. The child first grasps that two points are connected, that one region encl... ai Sep 11 Caption-once, Frames-on-Demand: Visual-Need Routing for Budget-Aware Agentic Long Video Understanding Long video understanding on edge devices runs into a fundamental tradeoff. You can subsample visual tokens and hope the model still catches what matters, or you can convert the video into text captions and lose the fine-grained vi...