Member login

Apply

AI Circle · About

A circle organized around evidence.

A research society for the people doing the operational work behind model capability: evaluations, post-training, human data, and infrastructure.

A letter from the founder

The floor is yours.

Albert Chun · Founder

Independent by design

Bootstrapped. No venture capital. No investors. Sponsors support rooms, not conclusions.

At AI Circle, we skip the grand theories of artificial intelligence and obsess over the operational weeds of model development: evaluations, post-training, and infrastructure; these are the parts that decide whether agents, enterprise deployments, and training environments actually work.

The obsession is in the name. A century ago the Vienna Circle held that a statement is meaningless unless you can verify it. The AI industry is full of statements that fail that test. We use words like capable, aligned, and reasoning as if they were physical properties. Without a codified evaluation behind them, they’re meaningless.

I’ve spent my career in human data, working on 100+ post-training projects that shaped the behavior of many then-SOTA models. Behind the scenes, when you’re scrutinizing rows of ideal outputs and preference rankings, models become less mysterious. Underneath the fanfare, frontier models are shaped by incredibly specific human feedback.

We partner with the human data and infrastructure companies that share this reality and are building the machinery to scale it. They sponsor many of our events. They don’t own our point of view. We’re bootstrapped, we’ve taken no venture capital, and we have no investors, because money with strings eventually turns you into a mouthpiece for the prevailing narrative. I built this room to strip away the marketing and talk honestly about what works and what fails.

The goal is to fuel the research that pushes AI forward. But every breakthrough has to be earned: taught through human data, verified by rigorous evaluations, and made operational by infrastructure that lets the research scale. That pipeline is the work that separates AI vaporware from reality.

Welcome to the room. The floor is yours.

Albert Chun
Founder · AI Circle

I.

Abstract

T

he Vienna Circle met for twelve years and left behind a standard rather than a doctrine. Say what would show you right or wrong, or you have said nothing at all.

We took the same standard for a simpler reason. A researcher will say the honest version of a result to twelve people in a room, and the safe version to twelve hundred online. We build rooms for the twelve.

II.

Chapters & institutions

AI Circle members connecting over drinksAI Circle San Francisco chapter mixerAI Circle Paris chapter dinner by candlelightAI Circle members in conversationAI Circle interview at a member eventAI Circle members at a daytime gatheringAI Circle members connecting over drinksAI Circle San Francisco chapter mixerAI Circle Paris chapter dinner by candlelightAI Circle members in conversationAI Circle interview at a member eventAI Circle members at a daytime gathering

San Francisco

37.7749° N, 122.4194° W

Seattle

47.6062° N, 122.3321° W

Toronto

43.6532° N, 79.3832° W

New York

40.7128° N, 74.0060° W

London

51.5072° N, 0.1276° W

Paris

48.8566° N, 2.3522° E

Stanford University

Campus circle

Massachusetts Institute of Technology

Campus circle

AI Circle members laughing togetherAI Circle San Francisco chapter conversationAI Circle poker night mixerAI Circle members networkingAI Circle at MACHINA conference stageAI Circle San Francisco chapter groupAI Circle members laughing togetherAI Circle San Francisco chapter conversationAI Circle poker night mixerAI Circle members networkingAI Circle at MACHINA conference stageAI Circle San Francisco chapter group

The standard

Say what would show you wrong.

Capability claims matter only when they can be verified. The Circle exists to put unfinished claims in front of people capable of testing them.

Enter the room