Google's TabFM skips per-dataset training and still predicts on tables it's never seen

The vast majority of business data is tabular — living in data warehouses, CRMs, and financial ledgers — yet building a reliable model from it still means training a new one from scratch for every dataset, then maintaining hyperparameter tuning loops, feature engineering, and retraining pipelines to fight data drift. Google Research is proposing a way around that: a new foundation model called TabFM that treats tabular prediction as an in-context learning problem instead. It can generate predictions for a new, unseen table in a single forward pass.