Why the Most Advanced Models Do Not Always Deliver the Best Results in AI Projects
Research shows that a large portion of artificial intelligence projects could be cancelled due to costs and uncertainties.
Researchers at the Massachusetts Institute of Technology have revealed that the most expensive and capable models in artificial intelligence projects do not always deliver the best performance. Rising costs and uncertain business values are threatening projects.
The Future of AI Projects
It is projected that a significant portion of artificial intelligence projects will be cancelled by 2027 due to uncertain business value and soaring costs. Organizations are making critical decisions during this process.
Tested on Vendor Invoices
In a study conducted with a globally operating company, an artificial intelligence system that resolves errors in vendor invoices was tested. Different model configurations were compared during the tests performed on the system.
The Most Expensive Model Ranked Last
In the tests conducted, the most expensive and most capable premium configuration exhibited the worst performance. Budget-friendly and mid-tier configurations achieved a higher success rate.
Reasons for the Capacity Trap
Among the main reasons for the failure of high-end setups are overthinking, incompatible components, and unnecessary structural complexity. Matching the right model with the task is of great importance.