Causal methods for estimating individualized treatment effects in acute ischemic stroke from observational data.
Evaluates geospatial foundation model Clay for improving landslide detection via pixel-level segmentation.
Benchmark for evaluating AI agent capabilities in cybersecurity tasks within realistic multi-host cyber ranges.
Theoretical framework analyzing representation costs of parametric data-fitting methods through function spaces.
LLM-driven framework for generating interpretable clinical decision rules from tabular medical data.
Headroom compresses LLM inputs (tool outputs, logs, RAG chunks) by 60-95% with local-first algorithms, supporting AI agents.
Developer upgrades ishi, a Git embedding program using pgvector for semantic code intelligence, to Zig 0.16.
Interactive walkthrough of Andrej Karpathy's 200-line GPT implementation with visual breakdowns of fundamental concepts.
Cursor announces 1.5 trillion parameter coding model training on 100,000 GPUs.
Show HN: Opencode fork with stateful AI agent featuring autonomous memory management and integration with coding assistants.
SQLazy uses AI to generate complex SQL queries step-by-step in natural language with compiler verification and auditability.
Research paper showing pruned LLMs pass multiple choice benchmarks but fail on reasoning tasks, revealing benchmark limitations.
Policy analysis of U.S. Commerce Department export controls on Anthropic's AI models and lack of legal framework governing frontier AI restrictions.
Opinion essay on societal impacts and concerns around AI deployment.
Startup research tool using AI agents to create free alternative to Crunchbase by researching and profiling companies.
GitHub Copilot transitions to usage-based billing model for all plans after consuming included AI credits.
Research comparing AI agents vs humans on long-horizon tasks. Agents plateau within 24 hours while humans improve over weeks, showing limits of current test-time compute strategies.
LifeSciBench: Expert-written benchmark for evaluating agentic AI systems on complex real-world life science research tasks beyond simple fact-recall.
Open source tool for grammar and spelling correction using Apple's on-device Foundation Model, runs locally on Apple Silicon Macs via Swift.
Gartner report predicting 40% of organizations will demote or decommission AI agents due to governance and autonomy management challenges.
Terminal-based coding agent in Go compatible with DeepSeek V4, Qwen, local LLMs via OpenAI API, featuring vector memory and autonomous error recovery.
Newsletter tracking Ruby, Rails, and AI coding agents ecosystem, featuring RubyLLM integration for agentic applications.
Local proxy that reduces LLM token consumption by ~70% through context optimization without quality degradation, with A/B benchmarking validation.
Tokdiet local proxy reduces LLM token usage ~70% via context optimization with A/B testing validation across 66 tasks.
Explores whether gzip compression can function as a language model using compression-prediction equivalence.
OpenCoworker desktop AI agent built on aisuite enabling research, task automation, file access, and scheduled automations.
Tool predicting Claude API token costs and model selection before execution, with feedback loop for cost optimization.
Technical tutorial on GPU-accelerated matrix multiplication achieving 2678x speedup, relevant for ML infrastructure.
Research comparing general-purpose LLMs (GPT-5.2, Gemini, Claude) against specialized clinical AI tools on medical knowledge tasks.
PromptShark MITM proxy debugger for AI agents with loop detection, request logging, and real-time dashboard.
Python SDK (orkaia) providing guard decorators for function-level security or validation.
Claim about building fail-closed execution gate for AI agents. Title only, insufficient detail.
Qode autonomous coding agent generates 50k-line codebases via terminal TUI, supports multiple LLM providers and MCP servers.
ClimateSOS open-source foundational charter for guardrails and reuse in AI/LLM climate and transition-planning tools.
Announcement of GLM-5.2 open weights frontier AI model with minimal details.
Anthropic pauses token-based billing increase for Claude Agent SDK that would have raised heavy user costs.
GitHub retiring GitHub Models service; existing customers can continue using playground and API, new customers cannot access.
Privatemode platform using confidential computing to encrypt data before processing, keeping it protected during AI operations.
Local web dashboard to find, overview and manage Claude Code sessions across your machine with list, timeline, and kanban views.
Bayer's PRINCE platform: production agentic RAG system for pharmaceutical research using Text-to-SQL to query safety study data.
NewsGuard audit finding Mistral's Le Chat repeated state-sponsored disinformation 50-56.6% of the time when prompted.
SAMF framework providing machine-readable validation contracts for LLMs using MoSCoW prioritization for agentic systems.
macOS menu-bar app using Apple Vision framework to detect nail-biting habits via webcam with on-device processing.
cuTile Rust enables safe, data-race-free GPU kernels using Rust's ownership model with support for sync/async launches and CUDA graphs.
Guide to optimizing FP8 GEMM kernels on AMD Instinct MI355X GPUs using Matrix Cores in HIP.
Ril is a parallel data streaming tool for Python that distributes CPU-bound work across cores using Arrow RecordBatches.
Dopamine is an open-source cross-platform effects library built by and for AI agents with configurable intensity/mood parameters.
ATOM framework optimizes LLM inference on AMD Instinct GPUs for high-concurrency, long-context workloads with distributed scaling.
Rapunzel is a browser-style UI for managing multiple AI agents across sessions, inspired by tree-style tabs.
Terminal UI for Azure DevOps enabling PR, work item, and pipeline management from command line.