Nat Friedman's AI Experiments

Table of content
Nat Friedman's AI Experiments

Nat Friedman is a technologist and entrepreneur who approaches AI with relentless curiosity and speed. After serving as CEO of GitHub from 2018-2021, he’s been building personal AI experiments, funding AI startups through AI Grant, and leading the Vesuvius Challenge—a $1.5M+ prize competition using ML to read ancient Roman scrolls carbonized by Mount Vesuvius.

Background

Philosophy: Go Fast

From his personal site, Nat’s core beliefs about building:

“It’s important to do things fast. You learn more per unit time because you make contact with reality more frequently. Going fast makes you focus on what’s important; there’s no time for bullshit. ‘Slow is fake.’”

This manifests in his projects—rapid experiments that explore AI capabilities rather than polished products.

natbot: Browser Automation with GPT-3

natbot (1.9k stars) was an early experiment in LLM-powered browser control, released in late 2022. The concept: give GPT-3 a serialized view of the DOM and let it drive a browser through natural language commands.

From the README:

“Drive a browser with GPT-3”

The project outlined future improvements:

natbot pioneered ideas that later appeared in more sophisticated agent frameworks. Its simplicity made it influential—one Python file demonstrating that LLMs could reason about web interfaces.

openplayground: Local LLM Exploration

openplayground (6.4k stars) is “an LLM playground you can run on your laptop.” It provides a unified interface for comparing different models:

This reflects Nat’s philosophy of personal AI tools—run it locally, experiment freely, learn through direct interaction.

Vesuvius Challenge: ML for Ancient Texts

Perhaps Nat’s most ambitious AI project is the Vesuvius Challenge, co-founded with Daniel Gross and Dr. Brent Seales.

The goal: use machine learning to read the Herculaneum Papyri—ancient Roman scrolls carbonized when Mount Vesuvius erupted in 79 AD. These scrolls have been unreadable for 275 years.

MilestoneAchievement
March 2023Challenge launched with $1M+ in prizes
2023Grand Prize claimed—first text recovered
2024Multiple scrolls being decoded
2026$1.5M+ awarded, ongoing progress

The technical pipeline combines:

Sponsors include the Musk Foundation, Patrick & John Collison, and Matt Mullenweg.

AI Grant: Funding AI Builders

AI Grant is an accelerator Nat co-runs that provides:

Notable advisors: Andrej Karpathy, Patrick Collison, Tobi Lütke, Guillermo Rauch, Dylan Field.

AI Grant has funded projects like:

Other Projects

Key Principles

From nat.org:

PrincipleImplication
“Time is the denominator”Speed of iteration > perfection
“Smaller teams are better”Faster decisions, no room for mediocrity
“The efficient market hypothesis is a lie”Opportunities exist where others don’t look
“Where do you get your dopamine?”Better from improving ideas than validating them
“You can do more than you think”We’re tied down by invisible orthodoxy

Why It Matters

Nat represents a particular approach to personal AI: rapid experimentation, open-source sharing, and funding others to explore. His projects aren’t products—they’re probes into what’s possible.

natbot showed that LLMs could control browsers before agent frameworks existed. openplayground gave developers a local sandbox when most were stuck with OpenAI’s playground. The Vesuvius Challenge proved that ambitious, seemingly impossible problems could attract world-class ML talent when properly incentivized.

Next: Simon Willison’s LLM Tools

Topics: ai-tools browser-automation llm-playground machine-learning open-source