PACED: LLM distillation method that weights training problems by student competence using gradient signal-to-noise ratio to improve distillation efficiency.
Framework addressing causal confusion in end-to-end autonomous driving models through causal intervention during training to improve reliability and safety.
Methodology for detecting prompt injection across multi-agent LLM pipelines. Stage-level kill-chain tracking for attack resilience evaluation.
Detection and mitigation of object hallucinations in vision-language models. Bayesian approach analyzing attention weights and token confounders.
3D Gaussian splatting for weather prediction downscaling. Proposes scale-aware vision transformer for arbitrary-resolution atmospheric forecasting.
Training-free semantic segmentation using vision-language models. Global context-aware framework for dense prediction without additional training.
Experiment using Claude to autonomously build a website designed to generate traffic, exploring AI agent capabilities and decision-making in open-ended tasks.
MCP server enabling long-term memory for LLMs using SQLite, hybrid search (BM25+vectors), and local embeddings without API keys.
Live leaderboard comparing AI model subscriptions and API pricing across 27 benchmarked models from Claude, GPT, Gemini, DeepSeek, and others.
Multi-agent framework with persistent memory across sessions where agents collaborate on shared codebases and retain conversation context.
Case study documenting indecisiveness in AI coding agent using Claude Opus 4.6 when debugging non-trivial bugs in GoAWK.
Error tracking tool designed specifically for AI agents with CLI interface, compatible with Sentry SDK for existing setups.
30-day experiment running autonomous AI system with memory and sleep cycles, documenting emergent behaviors and their implications.
macOS/iOS app automatically redacting sensitive personal, financial data, faces, and metadata before sharing documents with Claude and ChatGPT.
arXiv paper on Springdrift framework providing auditable persistent runtime environment for LLM agents.
Enterprise architecture analysis on three-layer collapse in business process automation systems, discussing MCP servers and small LLM deployment.
Analysis of exposed Claude Code source revealing engineering practices: 259 PRs, 497 commits, 40K lines in 30 days, examining AI-assisted development culture.
Posse is a web UI for Anthropic's Managed Agents, providing browser-based interface for agent creation, sessions, and memory management.
Stork.AI is a directory of 14k MCP servers and AI tools with community trust scores, offering a meta-MCP server for discovering integrations within Claude, Cursor, and other IDEs.
Entroly is a context compression engine that reduces LLM API costs by 80% for Claude, Cursor, and OpenAI by compressing codebase context without losing visibility.
Research on using distributed AI agents with independent context windows to improve reasoning on complex multi-perspective questions.
NeonD is an open-source Postgres control plane based on NeonDB architecture with branching and PITR support.
Opinion piece comparing AI adoption to TV, discussing shift to AI-assisted programming and loss of challenging side projects.
Anthropic's 2023 statement on AI safety risks and impact, discussing concerns about powerful AI development in coming decade.
Buildermark open-source tool measures code generation by AI agents vs human developers by matching agent conversations to git commits.
Article on training AI robots to understand physical movement through human trainers in India and global industrial settings.
Tool testing brand visibility across ChatGPT, Gemini, Claude, and other LLMs when users ask buying questions, with competitive analysis.
GrimmBot: Autonomous AI agent in sandboxed Docker with desktop/browser control, self-improvement capability, and tool building.
1-bit quantized GPT with 800K parameters runs inference in 8KB of SRAM, demonstrating extreme model compression.
Lumisift: Open-source tool improving data retention in RAG pipelines from 40% to 87% by fixing retrieval accuracy for scientific documents.
CLI tool generating AI configuration files for Claude Code, GitHub Copilot, and Cursor from unified templates.
LLM-Wiki: Obsidian-based persistent agent memory system with searchable indexed documents, inspired by Andrej Karpathy's concept.
Agentjail: Minimal Linux sandbox for running untrusted AI agent code, with DNS filtering and optional GPU passthrough support.
is.team: AI-native project management platform with built-in AI agents as teammates, workflow automation, and external agent integration.
Local LLM server supporting Intel NPU and ARC GPUs with OpenAI/Ollama-compatible APIs, auto-detecting hardware.
Proactive AI agent on iMessage with 135+ tools that anticipates user needs by integrating email, calendar, and other services.
TUI diff viewer tool designed for AI agents to review code changes inline with annotations, enabling feedback loops without leaving terminal.
Claude Code skill using cryptographic randomness and Tarot cards to resolve ambiguous planning decisions through entropy injection.
Startup built cloud ops AI agent 3 years ago enabling plain English queries across AWS, GCP, observability tools; concerns about competition from managed agents.
Specsight: Tool auto-generating living product documentation from codebase to reduce knowledge gaps and interruptions.
Styxx: Python package providing real-time monitoring of LLM agent cognitive states through token probability analysis and reasoning detection.
Microsoft executive suggests AI agents may need software licenses; discusses industry response to AI threats from Salesforce and others.
Essay exploring whether LLM failures stem from fundamental architectural issues or insufficient scale, discussing scene completion examples.
Review of WordPress 7.0 release delayed due to real-time collaboration data storage issues, noting underreported useful features.
Kon is a minimal terminal coding agent with a 270-token core prompt, 6 default tools, and layered project-specific context for extensibility.
AndesCode: local offline AI coding assistant running Gemma 4 26B that indexes codebases and answers questions without data leaving machine.
Project Glasswing: analysis of how advanced AI models discovered in late 2025 excelled at identifying zero-day software vulnerabilities.
Diff4: tool converting AI agent file changes into shareable encrypted web preview links, compatible with OpenClaw and Hermes agents.
Browser extension using AI to filter Twitter/X feed by user-defined topics like crypto and rage politics with local model caching.
Personal account of how AI coding assistants changed development workflow and pace, shifting from manual linear processes to iterative exploration.