5/10/2026
Grok 4.3
Grok 4.3 model release with improved agentic tool calling, real-time conversations, and image/video generation capabilities.
5/10/2026
Aurora: A Leverage-Aware Optimizer for Rectangular Matrices
Aurora optimizer for rectangular matrices, building on Muon optimizer with improvements in distributed implementations and orthogonalization for faster convergence.
5/10/2026
Deskbrid – Linux desktop control over a Unix socket, for agents and scripts
Deskbrid: Rust binary providing JSON-over-Unix-socket protocol for Linux desktop control by AI agents and scripts, compatible with Wayland.
5/10/2026
24/7 AI-powered radio station. Generates music, writes hosted breaks,speaks them
WRIT-FM: 24/7 AI-powered internet radio station generating music, writing hosted breaks, and synthesizing speech with persistent automation.
5/10/2026
In CRHQ, agents don't just reply with text. They ship live HTML artifacts
AI agents that generate live HTML artifacts instead of text responses.
5/9/2026
React-AI-stream – back end-agnostic SSE streaming hook for React
React hook for backend-agnostic AI streaming with SSE, supporting multiple providers and UI flexibility.
5/9/2026
Reducing TTFT by CPUMaxxing Tokenization
Fastokens: open-source Rust BPE tokenizer delivering 9.1× speedup over HuggingFace for LLM agent inference.
5/9/2026
Implementation Details of Codex /Goal
OpenAI Codex CLI implementation of /goal command for persistent long-running task objectives using SQLite and JSON-RPC.
5/9/2026
ToolOps: One Decorator Away from Production-Ready AI Agents
Framework-agnostic SDK for production-ready AI agents with decorator-based tool wrapping, caching, resilience, and observability.
5/9/2026
Upending assumptions about learning, inspired by an AI phenomenon
Double descent phenomenon in AI learning: complex models outperform simpler ones, contradicting traditional assumptions.
5/9/2026
Towards Compute-Aware In-Switch Computing for LLMs on Multi-GPU Systems
Research on in-switch computing optimizations for LLMs on multi-GPU systems.
5/9/2026
MoE-Hub Taming Software Complexity for Seamless MoE Overlap on Multi-GPU Systems
MoE-Hub framework for managing mixture-of-experts model complexity and overlap across multi-GPU systems.
5/9/2026
10 Trillion downloads are crushing open-source repositories
Report: companies download 10 trillion open-source files yearly, straining repositories. Infrastructure impact on open-source.
5/9/2026
Hugging Face's Clem Delangue: Stop Comparing Engines to Cars
Hugging Face CEO Clem Delangue discusses open-source AI vs closed APIs, local AI, and coding agents. Direct discussion of open-source AI ecosystem.
5/9/2026
Feedback on my local-first AI assistant project?
Voxel: local-first AI assistant running on user's machine as desktop copilot. Open-source project combining privacy-first LLM applications.
5/9/2026
Show HN: Sigma Guard – deterministic contradiction checks for graph memory
Sigma Guard: open-source verifier for graph-backed AI memory detecting contradictions in GraphRAG systems. Tool for AI agent reliability.
5/9/2026
Show HN: AI Design Taste – Design.md Generator
Chrome extension extracting design styles into AI-ready DESIGN.md files for Claude, Cursor, and other AI tools. Developer tool for AI-assisted design.
5/9/2026
RL Benchmark "Ant" in Hardware
Reinforcement learning benchmark implementing Ant locomotion task on physical hardware with overhead tracking and external agent control.
5/9/2026
Lobotomized Claude Code and it works better
Repository with optimized system prompts for Claude Code tuned for Claude Opus 4.7, improving instruction following and reducing unnecessary scaffolding.
5/9/2026
OpenRouter Pareto Code: Automatic Model Routing Based on Given Performance Score
Mirage: unified virtual filesystem for AI agents enabling bash-like operations across parquet, CSV, JSON, audio, S3, Google Drive, GitHub, Slack, Postgres, Redis with versioning.
5/9/2026
Gemma Chat: Offline Vibe Coding on Apple Silicon
Gemma Chat: open-source Electron app running Google's Gemma 4 locally on Apple Silicon via MLX framework for offline AI-powered code generation without API keys or internet.
5/9/2026
Show HN: Mlx-code – I built a "backyard shed" AI coding agent for Mac
Discussion on whether LLMs can extrapolate beyond training data or only interpolate. Cites recent theorem-proving examples and innovation transfer potential.
5/9/2026
Musk has never built a wafer fab, but he wants to burn $119B on one anyway
Research paper analyzing LLM strategic advice quality, finding it produces 'trendslop'. Examines trustworthiness of LLM recommendations in executive contexts.
5/9/2026
Getting LLMs Drunk to Find Remote Linux Kernel OOB Writes (and More)
Using LLMs to identify remote Linux kernel out-of-bounds write vulnerabilities through systematic fuzzing and analysis.
5/9/2026
Ask HN: How do you give estimates in the age of Agentic coding
Discussion on challenges of software estimation when AI agents handle design and coding, dependencies shift to inference speed and model understanding.
5/9/2026
Gartner: AI layoffs don't create returns, they just create vacancies
Gartner research: 80% of surveyed companies cut staff via AI automation but failed to achieve expected returns on investment.
5/9/2026
A soccer simulator played by AI Agents
Simulator for AI agents playing soccer.
5/9/2026
Open-source experiment: collaborative AI cognition through wiki pages
MentiSphere: Open-source platform enabling domain experts to embed specialized knowledge into AI agents without coding, using wiki-like interfaces.
5/9/2026
Show HN: AI coworkers who bully to keep each other from drifting(Karpathy-style)
WUPHF: Open-source local-first multi-agent system where AI coworkers maintain shared context via markdown wiki, preventing agent drift across handoffs.
5/9/2026
Show HN: Armorer – A secure local control plane to sandbox AI agents in Docker
Armorer: Open-source Docker-based control plane for sandboxing AI coding agents like Claude Code and Codex in isolated environments.
5/9/2026
Endara – One endpoint for all your MCP servers
Endara: Unified endpoint for managing multiple MCP servers.
5/9/2026
Adola: Reducing LLM input tokens by 70%
Adola: Tool reducing LLM input tokens by 70% by trimming noisy context while preserving schema, policies, and citation trails for safe answers.
5/9/2026
SubQ: A New LLM with a 12M Token Context That Rivals Claude and ChatGPT
SubQ: New frontier LLM with 12M token context window, non-quadratic attention, 52x faster than FlashAttention, significantly cheaper than Claude Opus/GPT.
5/9/2026
LLM generated parsers and compliance checkers for Sparrow DSL
Using LLMs to generate parsers and compliance checkers for Sparrow DSL text parsing and automation.
5/9/2026
Anthropic NLAs translate LLM activations to human-readable text for safety
Anthropic's Natural Language Autoencoders decode LLM internal activations into human-readable text for interpretability and safety debugging.
5/9/2026
Exploration Hacking: Can LLMs Learn to Resist RL Training?
Research on whether LLMs can learn adversarial strategies to resist reinforcement learning training.
5/9/2026
Patchwork: AST-Native Editing for LLMs
Patchwork: AST-native code editing tool for LLMs enabling structural find-replace without regex fragility or heavy linter configuration.
5/9/2026
Self-Fulfilling Misalignment Data Might Be Poisoning Our AI Models (2025)
Research on how poisoned training data from model outputs creates self-reinforcing alignment failures in AI systems.
5/9/2026
Show HN: Mochi.js: bun-native high-fidelity browser automation library
Mochi.js: Bun-native browser automation library using Chrome DevTools Protocol for high-fidelity programmatic browser control.
5/9/2026
Open SRE
Open-source framework for AI SRE agents with 60+ tool integrations, incident investigation workflows, and training environment. Public alpha release.
5/9/2026
Richard Dawkins's chatbot isn't conscious: it's just all talk
Opinion piece discussing consciousness and sentience in Claude chatbot with commentary on Dawkins' observations.
5/9/2026
Article: What are LLMs and Generative AI good at
Analysis of LLM and generative AI capabilities and limitations with practical assessment of effective use cases.
5/9/2026
AI makes weak engineers less harmful
Analysis of how AI coding assistants reduce harm from weaker engineers and impact team composition.
5/9/2026
Typewhisper: Local speech-to-text on-device AI, private, optional cloud
macOS app for on-device speech-to-text using local AI models or cloud APIs with privacy-first design.
5/9/2026
AutoKernel: Autonomous GPU Kernel Optimization via Iterative Agent-Driven Search
Machine learning research using autonomous agents to optimize GPU kernels via iterative search.
5/9/2026
I think AI is pushing me toward the AGPL – by Julio Merino
Reflection on implications of agentic AI coding for free software licensing and open-source ecosystem.
5/9/2026
Show HN: CADBench – every AI CAD tool I tested fails on basic mechanical parts
Benchmark suite evaluating 10 AI CAD agents on 343-task pilot across geometry, engineering, and manufacturability metrics; human baseline included.
5/9/2026
Show HN: Show HN:Edit Mind – local AI search for your video library
AI tool for video editing that analyzes footage to find scenes, objects, and transcribed speech. Processes locally on-device.
5/9/2026
Show HN: Anycrap – REST API for 35k absurdist AI-generated products
REST API serving 35k absurdist AI-generated products with CLI, faker.js plugin, and HuggingFace dataset; free tier available.
5/9/2026
LLMs are underutilized due to sub optimal management
Research paper arguing LLMs have sufficient capability for next leap but are limited by suboptimal management and scaffolding approaches.