It’s not you, it’s your AI
There have been a series of papers published recently that showed that large language models have their own biases when making investment recommendations.
Recent studies have revealed that artificial intelligence models possess their own biases when offering investment suggestions. In a recent paper, researchers from South Korea and the United States have demonstrated that this bias can be manipulated within an open-weight large language model by adjusting a single neuron in the neural network.
The study focused on five widely used models: Qwen3, Llama 4, DeepSeek R1, Gemma 4, and Mistral Small. By altering the activation of a specific neuron in each model, the researchers were able to modify the investment recommendations for individual stocks or entire portfolios without altering the initial prompt or modifying any model parameters.
This finding is particularly concerning, as it suggests that all AI models have inherent biases, which could lead to biased investment advice for users. In closed-weight models, such as the frontier models, advisers are unaware of the model's bias and cannot adjust the activation parameter to change the recommendation. However, in open-weight models, the activation parameter can be identified and intentionally manipulated, allowing advisers to tailor the model's investment bias to suit their preferences.
Written by urgent.news from Klement on Investing's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
