42AI launches with $62.5M to predict how traders act when markets turn volatile

Israeli artificial intelligence startup 42AI announced its launch with $62.5 million in funding to develop a behavioral world model. This model aims to predict how individual traders will act during volatile market conditions. The company's software estimates the likelihood of each decision a trader might make in a specific situation.

Amit Gal-Or, co-founder and CEO of 42AI, explained that behavior is probabilistic and not deterministic. The company transforms knowledge about an individual into mathematical representations of personality traits, which a transformer-based world model then uses to calculate probabilities. A separate language model then documents the traits and context that drove each prediction.

Brokerages can utilize the model to anticipate each client's trading behavior over the next 24 hours, including the instrument, direction, and volume. If multiple clients appear set to engage in the same trade, a risk desk can hedge before the potential exposure affects their account. The model can also assess clients based on factors such as their frequency of calls and the sentiment in their voice.

Currently, 42AI has onboarding agreements with ten large brokerages, charging either a flat fee or a percentage of the value generated. The startup has already achieved seven figures in revenue, with one brokerage reportedly generating an additional $30 million through dynamic hedging. The model reportedly boasts a 98% accuracy rate in predicting trades, though this figure measures whether trading flows fall within a predicted range rather than the correctness of individual trades.

The 42AI team spent three years conducting research, utilizing data from 30 million people. Their Large Behavioral Model, described in a February paper, adapts Meta Platforms Inc.'s Llama 3.1, an 8 billion-parameter transformer-based model, to predict strategic choices derived from an extensive psychometric profile. Concerns about privacy were addressed, with Gal-Or stating that the system does not require personally identifiable information and that the model never knows the individual's identity.

The team built profiles for each person based on linked account activity. They also noted that brokerage companies could potentially exploit these predictions, but emphasized that internal ethics committees and regulatory oversight would prevent such misuse. Looking forward, 42AI is exploring markets such as gaming, fraud prevention, and cybersecurity, with a long-term objective of enabling robots and AI agents to understand human behavior.

The model's accuracy may decrease when external influences, such as advice from colleagues, are not accounted for.