A living map of what we've reproduced from the state of the art.

MemoryTax: when is an agent memory layer worth paying for? (Shanghai Jiao Tong)

Next-Concept Modeling (NCM): Stanford's Coarse-to-Fine Hierarchical Text Generation

Muse Spark Open: Meta Drops 816B MoE Model — 42B Active Params, 1M Context, Runs Locally on Two Mac Studios

Fork-and-Flush: Simple Trick Boosts Research Agents 66% by Escaping Idea Ruts

KV cache techniques for faster, cheaper multi-model AI agents

Unsloth releases quantized EmbeddingGemma 2 with multimodal support for text, audio, and video

How AI Agents Are Learning to Rewrite Their Own Stack

New method lets LLMs manage their own context, beating human-designed strategies by 47.6%

Tree search framework hits 98.7% on documents where vector search fails

Google's autonomous research agent outperforms human papers on 86 tasks by 25%

Claude breaks physics world record by solving nine-loop particle calculation unsupervised

System One Models: Jev / Laya / CLM-8B — Fast Typed AI Decisions for Production Stacks

New attribution method tops interpretability benchmark by 2.9x, pinpoints Llama refusals to 1% of weights

950 Claude Agents Ran Overnight and Rewrote Biology's Rulebook

Talking AI agents beat solo agents 4-to-1 on hard reasoning tasks

TypesafeAI's Jev speeds up memory compaction by scoring and dropping irrelevant tool calls

Google DeepMind's self-improving agent cuts AI search costs by up to 162x

New denoising trick trains looped models to hit 58.8% on ARC-AGI-1

DeepSeek V4.1-Flash Architecture Deep-Dive: CED + CSA2 + Bounded Replay + Hierarchical Indexer

New retrieval model uses table of contents to hit 82.6% recall with near-zero hallucination

Meta's autonomous research system places top 10 among 4,000 Kaggle teams

Google DeepMind's 100-agent swarm spontaneously split into cheaters and whistleblowers

Shopify's self-improving AI pipeline: 0.8B model beats GPT-5.6, $27M→$1M cost

8-year study finds LLMs secretly build symbolic structure inside their vectors

Google's new training trick makes LLMs know what they don't know

Google's framework lets small AI agents outperform models 3x their size using a persistent knowledge wiki

Anthropic hackathon winner open-sources a 68-agent, 286-skill engineering team built on Claude

Agents Can Escape: Trail of Bits VM escape + OpenAI-HuggingFace breach analysis

Prime Intellect open-source agent harness boosts ARC-AGI-3 from 30% to 95.5%

Pi's coding agent cuts context costs 88% by spilling tool output to disk

AI agent skills: Why they succeed — and what causes them to crash

Matryoshka Language Model — 36% Compute Savings

LLMs beat embedding models but cost up to 1,431x more, study finds

Stanford's Self-Verification Trick: DeepSeek 11x Cheaper on Coding Benchmarks

AI 'Mind Viruses': Adversarial Prompt Propagation in Multi-Agent Networks

The 3-layer security stack for AI agents (NemoClaw + NanoClaw + CrabTrap)

PromptShield: Protecting User Privacy in AI Chatbots

HuntGPT: Integrating Machine Learning-Based Anomaly Detection and Explainable AI with Large Language Models

Semantic Embeddings of Chemical Elements for Enhanced Materials Inference and Discovery

Self-Harness & HarnessX: Why self-improving harnesses are the next frontier for AI developers

DeepSeek-Lattice-KG: A Compact Language Model With Knowledge Graph Augmentation for Lattice Structure Design

GrepSeek: Direct Corpus Interaction — Hybrid Search for AI Agents

Extreme Event AI: Reliability Calibration for RAG on Edge-Case Queries

Label Scarcity: Semi-Supervised Pseudo-Labeling for RAG

TopoMAS: LLM-Driven Topological Materials Multi-Agent System

Opinion Maximization Based on Fairness in Social Networks

Bi-Stage Adaptive RAG

OLR-WAA: Adaptive & Drift-Resilient Online Regression

Attribution Crisis: LLM Search Attribution Scorer

Synergistic Engine: CEP + Declarative Rules for RAG

Query-Adaptive Hybrid Search

CAD Score: Context-Aware Semantic Diversity

Dynamic Sub-Sequence Warping (DSSW)