Keystroke timing patterns for authorship verification to detect AI-generated text, using 136M keystroke events to identify cognitive signatures.
Analysis of catastrophic outcomes from misspecified AI objectives, studying how reward hacking can lead to undesirable results.
Unified taxonomy with 11 dimensions for categorizing deep learning-based multivariate time series anomaly detection methods.
Multi-modal language model-based agent for vector sketch generation trained with process-reward RL. Includes ControlSketch-Part dataset with part-level annotations.
Method for quantifying self-awareness in continual robot learning by isolating invariant cognitive structures from rapidly acquired skills.
Bolzano, open-source multi-agent LLM system for mathematical research. Reports 8 problems solved with LLM assistance orchestrating prover and verifier agents.
Deep quantile process regression method for off-policy evaluation that estimates full return distributions rather than point expectations.
MARBERT model applied to emoji prediction in Arabic tweets using corpus of 11,379 tweets classified into 14 categories.
Symphony is an open-source orchestration spec enabling code generation workflows. Team built repo with zero human-written code using Codex and agent-friendly architecture.
Choco uses OpenAI APIs and AI agents to automate food distribution, processing 8.8M+ orders annually and reducing manual work by 50%.
MCP server for LinkedIn Marketing API enabling Claude to query campaigns, analytics, and Lead Gen Forms via 19 tools. Open source developer tool.
Self-directed learning curriculum for AI covering SOTA concepts. Educational resource curating modern AI knowledge.
WordPress integrating AI features as agentic operating system. Discussion on cost implications and infrastructure shipping. AI agents context.
Claude Code skill integrating evidence-based learning science into agentic coding workflows via adaptive exercises using spaced repetition and retrieval practice.
WaveletLM: attention-free transformer alternative using wavelet decomposition and FWHT achieving O(n log n) scaling. Outperforms GPT-2 Medium on WikiText-103.
Port of 11 Claude Code skills to OpenCode framework, including code review, security audit, and feature development. Installable via command line.
Primus projection tool estimates memory usage and training performance for distributed ML runs before execution.
MIT-licensed plugin extending Claude Code to plan before coding by asking clarifying questions and reducing verbose output.
On-device multimodal AI with real-time audio/vision conversation, multilingual support, runs locally. Open source LLM application.
Curated list of automations for Codex coding assistant with community contributions. Incomplete content.
Detailed technical analysis of fine-tuning Gemma 4 E2B with LoRA, examining changes to probability distributions and model behavior across 23 query types with 5K training pairs.
EPFL research enabling robots to learn from each other across different hardware via transfer learning. Robotics ML research.
Terminal AI coding agent with multi-agent runtime, long-term auditable memory, file/command execution. Python-based developer tool for local repos.
Microsoft Work IQ MCP Server and CLI for GitHub Copilot to access Microsoft 365 data and extend AI assistants.
Analysis of LLM behavioral degradation in long sessions showing sycophancy, hedging, and abandonment of previous positions. RLHF training effects.
GAI: Open-source Go library for building LLM agent applications with provider abstraction, prompt helpers, and agentic workflows without heavy frameworks.
Technical walkthrough of reverse-engineering a Flutter Android app's license algorithm from AOT-compiled Dart binaries using binary analysis techniques.
Marketing content for AI automation platform offering outbound, content, SEO, and assistant workflows with 6K+ teams using 3,367 AI agents in production.
Superpowers is a software development methodology for coding agents with composable skills and structured task planning.
MinIO archived repository notice. S3-compatible object storage for ML/analytics workloads.
Opinion on software alignment as the bottleneck in AI-driven development, noting 10x coding speed hasn't translated to 10x software quality improvements.
Open-source ERP system prototype using chat as primary interface, handling sales, invoicing, inventory, accounting, and analytics with Docker deployment option.
ReadTube tool transcribes YouTube videos into readable text via LLM, periodically syncing subscribed channels into personal newsletter format.
Minimal context engine with streaming API for creating and comparing prompts with local LLMs via Ollama, featuring objective-based workflow design.
Benchmark comparison of Claude Opus 4.6 vs 4.7 across effort levels and prompt steering, minimal content provided.
Discussion on scarcity of consumer ChatGPT apps compared to business applications. Market observation.
Research on weak supervision for LLM reasoning with 8-256 examples, noisy labels, and proxy rewards across math, science, graph domains using Qwen/Llama models.
NARE: Framework converting LLM System 2 reasoning into deterministic System 1 Python scripts via semantic compression, achieving O(1) execution latency.
Novel efficient code editing mechanism using hash anchors, Myers diff, and single-token anchors for AI agent tooling. 60% cost reduction.
Local multi-agent runtime separating thinking from execution with worker boundaries. Open-source AI agent platform with explicit control and verification.
Three-year retrospective on AI in drug development. AlphaFold impact, BenevolentAI failures, data quality bottlenecks. Industry perspective.
Analysis of challenges deploying AI-generated apps: isolation, database/auth/deployment issues. Technical insights on AI coding tool limitations.
AI copilot application for Indian retail investors. Stock/mutual fund research and portfolio analysis tool.
Analysis of Apple's AI supplier strategy: Google, OpenAI, Anthropic plus in-house layers. Non-cash partnerships and monetization mechanisms.
CLI and protocol for shipping LLM-friendly library context. Alternative to MCP servers for improving LLM code generation accuracy.
Native macOS AI code editor running locally on Apple Silicon via MLX/Ollama. Offline, no API costs, code obfuscation before cloud LLM use.
Case study: booking platform replaced UI onboarding with AI-driven interface. Simplified complexity through conversational AI instead of explicit design.
Quantitative study of LLM hallucinations in counting tasks across GPT-5.3, Gemini 3, Claude Sonnet 4.6. Evaluates Knowledge Innovation System suppression effects.
Open-source AI agent platform with browser control, DNS discovery, permissions, rate limiting. Standardized protocol making websites AI-ready.
Stanislav Fort's research on vulnerability detection using small LLMs at scale via nano-analyzer tool on FreeBSD/OpenBSD kernels.