Andrej Karpathy

Joins Anthropic to lead recursive self-improvement team

Andrej Karpathy analysis finds 57 million US jobs at high risk from AI

Published the LLM-Wiki framework on GitHub, sparking widespread adoption

stars:

5000+

Karpathy states the AI industry must reconfigure for an agent-centric future where AI agents, not humans, are the primary customers.

Former Tesla AI director and OpenAI founding member made public statement about industry shift to agent-centric AI

previous role:

Tesla AI director

current role:

Industry commentator

Built 'Dobby the Elf Claw' smart home AI agent for unified natural language control

product:

Dobby the Elf Claw

function:

Consolidates control of lights, HVAC, shades, pool, and security

Argued that AI agent failures are 'skill issues' related to instructions and tooling, not model capability.

Briefly released and then deleted a tool scoring 342 job types for AI exposure, finding $3.7 trillion in annual wages at high risk.

amount:

$3.7T

Published results from autonomous AI research experiment showing AI can improve neural network training

Released open-source 'autoresearch' tool for autonomous ML experimentation

Released open-source AI research agent that autonomously runs ML experiments

lines of code:

630

Open-sourced an autonomous AI research agent capable of running ~100 unsupervised experiments overnight.

Developed a personal knowledge management system using LLM embeddings for 400K-word research base, bypassing traditional RAG.

Andrej Karpathy declared programming has become "unrecognizable" due to AI agent reliability breakthrough

Delivered 'Engineering's Phase Shift' talk at 2024 Asilomar AI Summit covering AI psychosis, AutoResearch, and distributed AI research