Pythonx project integrates Python runtime into Erlang VM. NGI0-sponsored work adds full Python support to Livebook notebook platform with reproducible environments via uv.
BrowseComp benchmark evaluates AI agents' web search and information retrieval capabilities across 1,266 questions.
GitHub Copilot issue where users requesting advanced models silently default to lower-tier options.
ClawJetty tool provides live progress tracking interface for AI agents executing long-running tasks.
Open-source AI trading agent combining LLMs for reasoning with deterministic scripts for market execution, live trading with $1000.
Guide covering LLM evaluation metrics and methodologies for assessing language model performance.
Oblien: cloud VMs with REST API for safely executing AI agent code and file operations, designed for tools like Cursor and Claude Code.
Obsidian AI: open-source platform for building, deploying, and orchestrating AI agents and multi-agent workflows with unified control plane.
Open-source resume and job application tool using LLM integrations from multiple providers for tailoring and job tracking.
Comparison of three methods (Ablation, Heretic, Obliteratus) for removing refusal behaviors from LLMs.
Societies of large language models collaborating through natural language interfaces to solve problems, inspired by Minsky's society of mind concept.
Research on how diffusion models memorize training data, measuring latent dimensionality collapse on low-dimensional manifolds as data becomes scarce.
Spiking neural networks with adaptive task-switching for resource-constrained multi-task reinforcement learning in autonomous agents.
Data-driven survey identifying 14,648 papers on LLM limitations from 2022-2025 using LLM-based classification.
System using multiple pre-trained models and consistency-based reasoning to handle distributional shifts in novel environments.
Interleaved online fine-tuning method enabling LLMs to acquire capabilities beyond base model through reinforcement learning.
Survey of differential privacy techniques from symbolic AI through LLMs, covering definitions and applications.
Yokai: benchmark environment for zero-shot coordination in cooperative AI with belief tracking over space and time.
Empirical study analyzing intra-modality and inter-modality dependencies in multi-modal learning benchmark datasets.
RADAR: dynamic routing system that selects optimal reasoning LLM models based on task difficulty and cost-performance tradeoffs.
KVTC: transform coder that compresses KV caches for efficient LLM inference storage and memory management.
CostNav benchmark evaluating physical AI agents on real-world economic metrics for autonomous delivery systems beyond task success.
Co-design framework jointly optimizing robot hand morphology and dexterous control policies across multiple embodiments.
Closed-loop drug discovery system using language models, property alignment, and strategic search for de novo ligand design.
Proprioceptive-Privileged contrastive learning framework for sample-efficient whole-body control in humanoid robots using reinforcement learning.
First-order analysis of how cross-entropy training reshapes transformer attention geometry for probabilistic reasoning.
Benchmarking study comparing graph neural networks against classical heuristics on hard constraint satisfaction problems.
Autonomous AI analysts built on LLMs analyzing same dataset independently, quantifying variability in research conclusions without human coordination.
Study of how LLMs develop early syntactic structures that persist as errors through training, using OPT model on BabyLM dataset.
CARE: Evidence-grounded agentic framework using visual language models for medical reasoning with explainability and clinical accountability.
RACAS: Framework for controlling diverse robotic platforms with a single agentic system through unified interface.
Systematic comparison of training objectives (cross-entropy, prototype, triplet, AP loss) for out-of-distribution detection in image classification models.
Pluk enables agentic data notebooks on personal databases with plain language queries, SQL workflows, and interactive dashboard generation running locally.
Research on executing programs inside transformer models with faster inference. Limited detail in abstract.
CLI tool for managing AI agent skills from GitHub repos and local directories. Skill discovery and linking via SKILL.md manifests.
MCPDome: open-source security gateway for AI agents enforcing authentication, authorization, rate limiting and injection detection on MCP servers.
Laminae is an open-source Rust SDK adding guardrails, safety, personality, and learning layers to LLM applications for production use.
SwarmClaw: Control plane for managing multiple OpenClaw agent instances. Supports 14+ LLM providers, multi-agent workflows, and chat integrations.
TensorFusion.AI: GPU virtualization and pooling solution for Kubernetes clusters. Infrastructure for optimizing AI compute resources.
Essay on UX design challenges for AI products in 2026, addressing AI fatigue and defining best practices.
K9 Audit: open-source tool capturing intent-execution audit trails for AI agents via causal five-tuples per step.
Helios: Open-source harness for autonomous ML research agents. Unrestricted agent execution framework for machine learning experiments.
Guide on agentic engineering patterns for using AI agents to improve code quality rather than reduce it.
Open source local-first AI agent framework supporting small models with optional cloud hybrid support. v1.1 release with multi-agent capabilities.
Open source middleware layer adding persistent identity, auditable safety, and governance for AI agents. 542 passing tests, production-ready.
AI-nexus: Open source tool that reduces token usage in Claude Code by selectively loading only 2-3 relevant rules/skills per prompt instead of all 50+.
Two-model epistemic probe (PROBE/MAP) to capture LLM constraint topology and decision states before token generation.
Genesis V66 research on emergent reasoning in neural architectures using GRPO training without format rewards. AI agent learning from minimal feedback.
Open-source web tool detecting accessibility issues in Jupyter Notebooks with remediation features. Developer tool but limited AI focus.
DAG MCP (Model Context Protocol) for Claude Code supporting complex task execution in agents.