The Sequence AI of the Week #891: Prompting a Spreadsheet : Inside Google’s TabFM for Tabular AI

2026-07-08 · Source: TheSequence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Intermediate, quick

Summary

Google Research recently introduced TabFM, a new foundation model designed for tabular classification and regression tasks. This model aims to revolutionize the traditional machine learning workflow by producing predictions on unseen tables in a single forward pass, eliminating the need for training, hyperparameter tuning, and extensive feature engineering. TabFM operates on an "in-context learning" principle, where the entire problem, including both training and test rows, is provided as a single large prompt. This approach allows the model to generate answers directly, simplifying the process significantly. TabFM builds upon the success of TimesFM, a time-series foundation model from the same Google team, indicating a consistent strategy for developing foundation models that streamline complex data analysis tasks.

Key takeaway

For Machine Learning Engineers struggling with the repetitive feature engineering and tuning cycles for tabular data, TabFM offers a significant paradigm shift. You should investigate TabFM's in-context learning approach to potentially reduce development time and operational overhead for classification and regression tasks. Consider piloting TabFM on new datasets to evaluate its performance against traditional gradient-boosted tree models, especially for rapid prototyping or scenarios where quick deployment is critical.

Key insights

TabFM introduces in-context learning for tabular data, bypassing traditional ML steps like training, tuning, and feature engineering for classification and regression.

Principles

• Foundation models generalize tabular data.
• In-context learning applies to spreadsheets.
• Pre-trained models reduce feature engineering.

Method

TabFM processes the complete tabular problem, including training and test rows, as a unified prompt to generate predictions in a single forward pass without explicit training or tuning.

In practice

• Apply to churn, fraud, credit risk.
• Streamline tabular ML pipelines.
• Test on new, unseen datasets.

Topics

• TabFM
• Foundation Models
• Tabular Data
• In-context Learning
• Machine Learning Workflow
• TimesFM

Best for: AI Engineer, Research Scientist, AI Scientist, Machine Learning Engineer, Data Scientist

Related on AIssential

• Introducing TabFM: A zero-shot foundation model for tabular data — TabFM applies zero-shot in-context learning to tabular data, eliminating manual tuning and feature engineering for classification and regression.
• Attention Is Not What You Think! — In-context learning for tabular data is row attention, a mechanism Random Forests have employed for 20 years.
• TabH2O: A Unified Foundation Model for Tabular Prediction — TabH2O unifies tabular classification and regression via in-context learning, improving training stability and noise robustness.
• TabPFN-MT: A Natively Multitask In-Context Learner for Tabular Data — TabPFN-MT extends Prior-Data Fitted Networks to natively support efficient multitask in-context learning for tabular data.
• Data Presentation Over Architecture: Resampling Strategies for Credit Risk Prediction with Tabular Foundation Models — Context construction significantly impacts Tabular Foundation Model performance in imbalanced credit risk prediction.
• LLMs on Tabular Data with Limited Semantics: Evidence from Industrial Car Retrofit Prediction — LLMs serve best as complementary components to strong tabular machine learning baselines, not replacements, for industrial data with limited semantics.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by TheSequence.