ResearcherCautiousTier 2

Andrej Karpathy

Independent Researcher & Educator, Independent (formerly Tesla, OpenAI)

The rare researcher who can build frontier AI AND explain it to normal people — ex-Tesla AI director, ex-OpenAI, and the best AI educator on YouTube.

Credentials

Former Senior Director of AI at Tesla (Autopilot), Founding member and former research scientist at OpenAI, PhD from Stanford (under Fei-Fei Li), YouTube educator with 1M+ subscribers, creator of nanoGPT and Eureka Labs

Why They Matter

Karpathy bridges the gap between frontier AI research and practical understanding better than anyone. His YouTube tutorials have taught more engineers to actually build AI than most university programmes. For business owners, he's the person to follow if you want to genuinely understand what AI can and can't do — no hype, no doom, just clear technical insight explained in plain language.

Positions

AI Timeline View

AI is progressing rapidly, but we're still in the early chapters. Current LLMs are impressive but limited — like early operating systems. The real breakthroughs are still ahead.

Safety Stance

Cautious

Key Beliefs

LLMs are a new computing paradigm — not just chatbots, but a new kind of operating system that will reshape how we interact with computers.

Talk: "The State of GPT" at Microsoft Build

AI education should be hands-on and practical. You learn AI by building, not by reading papers.

YouTube channel and Eureka Labs founding mission

Tesla's pure vision approach to self-driving (no LiDAR, no HD maps) is the correct long-term architecture.

Tesla AI Day presentations, 2021-2022

The best way to understand neural networks is to build them from scratch — character by character, token by token.

YouTube series "Neural Networks: Zero to Hero"

Controversial Take

At Tesla, championed the removal of radar and ultrasonic sensors from vehicles in favour of a pure vision approach to self-driving. Critics argued this made cars less safe; Karpathy and Tesla maintained that vision-only is how humans drive and is the correct long-term bet.

Track Record

How well have Andrej Karpathy's predictions held up?

Pure vision (camera-only) self-driving can match or exceed sensor-fusion approaches

Made: 2021

Tesla's FSD has improved significantly but full autonomous driving remains elusive. Other companies continue using LiDAR.

Too Early

Transformers will become the dominant architecture, replacing recurrent neural networks across most tasks

Made: 2017

As an early OpenAI researcher, Karpathy was among those who saw the Transformer's potential early. It now dominates NLP, vision, and beyond.

Right

LLMs will become a new kind of operating system / computing platform

Made: 2023

AI agents and LLM-powered workflows are emerging rapidly, though the full "LLM as OS" vision is still developing.

Partially Right

Key Quotes

“The hottest new programming language is English.”
X (Twitter) (2023-01)

“I think of LLMs as the kernel process of a new kind of operating system. They can orchestrate tools, memory, and many forms of computation.”
Talk at Microsoft Build: "The State of GPT" (2023-05)

“The best thing about neural networks is that they work. The worst thing about neural networks is that they work — because we don't fully understand why.”
Stanford CS231n lecture (2017-04)

“Don't be a hero. Don't reinvent everything from scratch. Start with known architectures, known hyperparameters, and iterate from there.”
"A Recipe for Training Neural Networks" blog post (2019-04)