Study revealed low accuracy of AI chatbots in financial matters

Edition Financial Times conducted a large-scale study showing that popular AI chatbots provide incorrect answers to financial queries in most cases. The analysis touched on the ability of models to interpret market data and provide correct recommendations to users.

During testing, experts asked neural networks questions related to analyzing company reports and forecasting financial indicators. The results showed that models often make factual errors, confuse numbers in balance sheets, and provide outdated data that can lead to losses when making investment decisions.

Such incidents highlight the critical problem of "hallucinations" in specialized domains. Despite the high speed of text generation, current LLM architectures do not have built-in mechanisms for verifying financial data, making their use as capital management advisors extremely risky.

Discussion of the study's results is actively being conducted in the Hacker News community, where users point to the lack of access to verified databases in real-time for models. Specialists note that without integration with reliable financial market APIs, the use of chatbots for asset analysis remains unsafe.

The study raises questions about current security standards when implementing AI in the banking sector and fintech services. Developers will have to solve the problem of ensuring the accuracy of answers before such tools can claim the role of full-fledged financial consultants.