New AI Betting by Fei-Fei Li and Andrej Karpathy: The Simulation Society

Every three months, public companies take questions from analysts during quarterly calls. But what if firms could predict these questions in advance and analyze their answers? That’s one of the skills mentioned by Simile, a new AI startup that came out of Stanford and is backed by renowned researcher Fei-Fei Li and OpenAI founder Andrej Karpathy.

Simile emerged yesterday (Feb. 12) funded with $100 million in a round led by Index Ventures. Alongside Li and Karpathy, the startup—which did not disclose its valuation—also counts investors including Quora founder Adam D’Angelo and Scott Belsky, a partner at A24 Films.

Li and Karpathy both have close ties to Simile’s founding team, which includes Stanford researchers Joon Park, Percy Liang and Michael Bernstein. Li is co-director of Stanford’s Human-Centered AI Institute and advised Karpathy during his Ph.D. study at university. He is widely known for seminal work such as ImageNet, a large image database that helped drive major breakthroughs in computer vision. Karpathy and Bernstein also contributed to that work.

Simile’s goal of using AI to model and model social behavior is entering unexplored research territory, according to Karpathy, who worked at OpenAI again Tesla before launching his education-focused AI startup. While large language models often present a single, unified personality, Karpathy says they are actually trained on data taken from large numbers of people. “Why not rely on that mathematical power: Why simulate one ‘person’ when you could be trying to simulate a population?” he wrote in a post to X.

That view supports Simile’s broader mission. The Palo Alto-based startup aims to simulate the real-world consequences of major decisions, from public policy to product launches, across virtual citizens that mimic human behavior. The team has tested this concept on a smaller scale with projects like Smallville, a 2023 Stanford experiment in which 25 autonomous AI agents interacted in a virtual environment.

Now, Simile is expanding the way he uses the business. After spending the past seven months developing its model, the company is already working with customers on applications ranging from product development to litigation prediction. CVS Health Corporation, for example, uses Simile to create simulated focus groups, while Gallup uses the platform to create digital polling panels. For phone calls, Simile can guess about 80 percent of the questions are analytical at the end ask, said Park, the startup’s CEO, during a recent appearance on TBPN.

Currently, Simile’s models are based on data from hundreds of thousands of people enrolled in its courses. Over time, the company hopes to expand that to a global scale of nearly 8 billion people.

Simile joins a growing wave of AI companies focused on using simulation to model real-world situations. Much of the research in this space focuses on virtual systems, such as robotics and autonomous vehicles, with “world model” platforms developed by firms such as Google again Nvidia.

One of the most prominent figures in the world of models is Li herself. In 2024, he took a leave of absence from Stanford to present World Labsa launcher that creates 3D digital environments from images and text instructions. The company has raised $230 million to date and is valued at over $1 billion.