Mirror Particle builds 'world model' to predict human behavior — SkimNews
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Mirror Particle, a 2-year-old San Francisco startup, is building what CEO Abhivyakti Ahuja calls a 'world model' built from scratch that simulates why humans do what they do and how behavior changes over time, rather than fine-tuning LLMs to role-play as target demographics.
- Abhivyakti Ahuja dismisses the dominant LLM approach as 'bringing a super soaker to Niagara Falls,' arguing that fine-tuning models trained on hundreds of billions of data points with small datasets leaves them 'stuck in the past' and unable to perceive the world as humans do.
- Mirror Particle has raised an angel round and says it's close to closing its first venture round; the company will compete in TechCrunch's Startup Battlefield 200 at Disrupt 2026 in San Francisco on October 13-15.
- The startup's model blends clients' customer data with current events, pop culture, and social media, focusing on 'revealed behavior' — what people actually do — rather than self-reported survey answers, and tracking how motivations shift as a demographic moves through experiences.
- In a pilot with a well-known pet food brand, Mirror Particle's engine found packaging imagery (chicken, beef, vegetables) didn't matter; the real sales ceiling was the brand's perception as mass market and cheap, a problem imagery alone couldn't fix.
- The startup enters a competitive field that includes Simile ($200 million raised at a $2 billion valuation), Aaru ($88 million at $1 billion), and Humans& ($480 million seed at $4.48 billion valuation, which launched Persimmon to model human behavior).
- Ahuja, originally from India, studied neuroscience and computer science at the University of Toronto under Geoffrey Hinton's influence, then worked at Amazon Robotics where she met co-founders Will Song and Thomson Yen.
Why it matters: Mirror Particle enters a market research and brand strategy sector crowded with well-funded rivals — Simile at $2B, Aaru at $1B, Humans& at $4.48B — betting that brands will pay for a 'general layer for anticipating human behavior' that replaces the standard LLM fine-tuning approach. If the model delivers 'revealed behavior' predictions plus the 'why' behind them, the immediate payoff is reallocated ad spend and product decisions for early brand customers.
Ask SkimNews
