Joon Sung Park
Related people
Episodes
Notes
Sourcery · The $2B AI Startup Simulating Human Behavior with 85% Accuracy
-
How Did Simile's Digital Twins Reach 85% Accuracy?
Joon Sung Park says agents built as digital twins of real people predicted behavior with 85% or better accuracy by the end of 2024. It was the first time the field could show that result, and it became the first building block for Simile.
The study took the Smallville architecture to 1,000 real people, a representative sample of the United States across gender, race, and ethnicity. An AI interviewer talked with them voice to voice about the story of a life, where they grew up, hard decisions, and who matters. Those interviews trained a twin of each person. The people left for two weeks, came back to a lab, and completed surveys and behavioral studies while the twins predicted what they woul
-
What Was the Smallville AI Experiment?
Joon Sung Park ran Smallville in 2022 and 2023 to see whether a language model could simulate a whole person, then a whole town.
The models were GPT-3 and then GPT-3.5. They were still rough, but training on the web, Wikipedia, and social media made it possible to pull out human behavior and generate new actions, not only write text or classify a task. Smallville was a game town in the style of Pokemon and The Sims, filled with 25 agents. Each agent had a name, a life story, and an architecture for memory, planning, and reflection. They woke up, went to work, and formed relationships. The simulated day was the day before Valentine's Day. One character organized a party, and others asked eac
-
Why Doesn't a Buyout Offer End Simile?
Joon Sung Park says a $15 billion offer to sell Simile does not end the work when he has every reason to believe an ambitious idea exists that only Simile can pursue. Simile will pursue that idea.
People who stay influential across eras treat the greatest prize for winning the current game as the right and invitation to play the next one. Extreme commitment, stamina, and the ability to keep getting excited are what carry them. The reverse can be a mode of failure: optimizing for one fundraising, one customer, or one best paper, winning it, and then losing interest. Institutions and individuals who achieved incredible things can disappear over time. He will not call every disappearance a fai
-
Is AI Behavior Simulation in Its GPT-4 Era?
Joon Sung Park says AI behavior simulation is in its GPT-4 and GPT-4.5 era. The technology is here to make an impact, and in about a year and a half to two years the simulation space will be where a coding agent as an industry is today.
The data and the model are extremely important, and a lot of the battle will also be in the market through a forward-deployed motion. In the recent agent-coding era, companies that were forward deployed found a lot of success. The relationship became stickier and much more defensible over time, and the same pattern will repeat in simulation. The technology is changing and evolving every single month.
-
Why Does Predicting Behavior Fail to Show What to Do?
Joon Sung Park says a forecast is not what people came for. A warning that a market will crash in two quarters does not help unless the aim is to beat the stock market. The next question is what to do now, and a correlation cannot supply that lever.
Asking the right questions matters. The interview idea came in 2023 and 2024, during podcasts about academic work at Stanford. After an hour or two, a good interviewer who asks the right questions knows a tremendous amount about a field and a person. Questions are both topical and general, and asking someone to tell the story of their life reveals pockets of knowledge. Behavior data is collected too, including transaction data and mobility data.
-
Are Universal Basic Income Simulations Worth Running?
Joon Sung Park calls universal basic income simulations good ones to run. He poses them as a way to show where the policy worked and where it did not, in place of a five-year study. The company is already delivering value, and a small part of him remains the researcher, drawn to questions that are societal and generational.
Thomas Schelling, a behavior economist in the 70s and 80s and one of his research heroes, applied simulation to racial segregation, the generational topic of that era, to understand its causal mechanisms. Climate change sits on the same scale. It is hard because of coordination, not only the physics. Many actors with competing incentives each optimize for their own goals
-
Why Did Winning Electric Vehicle Messaging Lower Sales?
Joon Sung Park says a message that looked like a clear winner for a new electric vehicle also shifted how people saw non-electric vehicles and lowered sales across that whole market.
The study ran before Simile. He worked with a few automobile companies, some of the largest players in the field, and electric vehicles sat at the top of every conversation: whether to do it, when, and how. One simulation tested how people would react to different messaging for the entry of a new electric vehicle, and how to optimize that messaging so the company could convey its values. One strategy came back as a clear winner and looked ready to go. The same messaging then shifted overall perception of the no
-
When Is a Simile Simulation Confident Enough to Act on?
Joon Sung Park says an AI simulation of a person is strong enough for a very high-stakes decision when confidence is higher than 90 percent, and usually when it is higher than 85 percent.
Users open a web interface on a SaaS platform, name who they want to talk to, and ask any number of questions. There is no limit, and the platform generates responses. The simulation is a new way to make decisions in a social setting, so the work with users includes change management: when to use a result, how to use it, where more data is needed, and how to validate it. Next to the foundation model that runs the simulation, a second model is trained only to predict accuracy. It outputs a probability from
-
Why Do We Live in a Simulation?
Joon Sung Park thinks we live in a simulation because any civilization will recreate itself to understand itself. People ask him often because of the technology he is building, and the experience still has to feel real and true to who we are.
Humans and societies like ours always want to understand themselves better. The best way to understand something is to recreate it. A simulation is a portrait of who we are as a society, and making one is an innate desire for a civilization like ours. Any civilization will create a simulation, and at some point we likely live in one.
A society rests on quantum building blocks of people's tastes and behaviors. Inherent characteristics might be among th