Forecasting AI gets better with more data, not just bigger models

Researchers built a new, massive dataset for testing AI models that predict future trends. It turns out, these models get much better when trained on more data, not just by making the models themselves larger.

Companies building AI models for things like finance or supply chains assumed that bigger models meant better predictions. This new benchmark shows that simply feeding models more real-world data makes a much bigger difference than just adding more parameters. This means developers can focus resources on data collection and curation rather than just scaling up model size, potentially making powerful forecasting tools more accessible and efficient.