A Four-Type Framework for LLM Wiki by karpathy
A developer building a personal knowledge base using Karpathy's LLM Wiki framework discovered that storing only declarative knowledge (facts and definitions) is insufficient for expert-level reasoning…
A developer building a personal knowledge base using Karpathy's LLM Wiki framework discovered that storing only declarative knowledge (facts and definitions) is insufficient for expert-level reasoning…
PostHog introduces 'scouts' as small autonomous agents that monitor product data and emit signals for action, forming a core part of its self-driving pipeline now in open beta. Scouts use markdown-bas…
An AI agent using the AutoResearch framework autonomously improved a small GPT's training recipe over 123 experiments on a single H100 GPU, achieving a best mean bits-per-byte (BPB) of 0.9774 ± 0.0019…
A developer argues that 'vibe coding' autonomy levels (L0-L5) are incomplete without a second axis: operator discipline, which measures how much work survives session boundaries as inspectable state. …
An experiment testing LLM councils found that they suffer from groupthink, retaining only about a quarter of unique, high-quality ideas from individual models while favoring consensus ideas. The peer-…
A developer at Microsoft describes how adopting a policy of automating any task done more than twice led to unexpected consequences, including increased context switching and errors, but ultimately en…
A developer built a character-level language model from scratch using Shakespeare's complete works, training a nanoGPT model on a consumer-grade MacBook Pro in about 15 minutes. The project demonstrat…
By early 2026, "vibe coding" — describing software in natural language and accepting AI-generated output — had become a mainstream development workflow, with Collins Dictionary naming it Word of the Y…
A developer built a personal wiki from their digital exhaust by processing 55 export directories into ~2,750 structured markdown pages. The project normalizes data from dozens of services into a unifi…
Andrej Karpathy's LLM Wiki design is being adopted as an external memory architecture for AI agents, mapping raw source documents into a structured markdown knowledge base. The system uses three layer…
A developer has created AGENTS.md, a structured file that serves as an operating contract for LLMs working within Obsidian vaults. Inspired by Andrej Karpathy's LLM Wiki concept, the system defines a …
Software engineers are shifting their focus from high-level abstractions to foundational computing skills, driven by the rapid advancement of agentic coding tools like Claude Code and Codex that outpe…
A developer created pluckmd, a CLI tool that extracts blog articles into markdown without per-site configuration, to enable AI agents to read and index web content. The tool automatically handles Java…
A developer released a minimal GPT implementation written entirely in Go, trained on Jules Verne novels. The model generates short text fragments like "Mysterious Island" and takes about 40 minutes to…
An AI agent deployed by PostHog during a Lisbon team offsite identified a three-year-old bug in the company's ClickHouse query engine that prevented timestamp filters from using the primary key correc…
Andrej Karpathy's LLM Wiki has evolved its search methodology to eliminate token waste caused by agents retrieving entire files for single relevant sections. The system now uses targeted retrieval to …
Anthropic held an onboarding day where a speaker was introduced in a manner compared to Michael Scott introducing Karpathy. The event's details were not accessible due to JavaScript being disabled in …
The article introduces LLM Wiki V3, a concept document that proposes a "compartmentalized wiki architecture" for scaling LLM-maintained knowledge systems. It argues that stateful LLM systems are built…
TokenZip v2 is a token compression engine that reduces LLM input token costs by up to 95% for coding copilots like Claude Code and Codex by transforming an entire codebase into a multi-level, queryabl…
`microgpt.py`, a minimal, dependency-free Python implementation of a GPT (Generative Pre-trained Transformer) model for training and inference. It includes a complete autograd system, a character-leve…