Event
Information society
Study: AI makes mistakes in 88% of complex financial tasks
1 sourcePublished · Updated
A study found that popular chatbots like ChatGPT make errors in an average of 57% of financial tasks and 88% in complex calculations. The best model, Claude Opus 5, also showed a high error rate, with one test suggesting a potential tax penalty of £17,500.
Key facts
- 57% error rate in financial tasks
- 88% error rate in complex calculations
- Claude Opus 5 makes mistakes in 39% of cases
- potential penalty of £17,500
- study conducted by Saturn
ChatGPT · Claude Opus 5 · Saturn
Sources1
- Independent
ForkLogEarliest tracked report
Discussion
Comments