Frank Hutter on TabPFN: tabular foundation model predicting in a single forward pass
Machine Learning Street Talk · youtube · 2026-09-24
Machine Learning Street Talk interviews Prior Labs co-founder Frank Hutter on TabPFN, a tabular foundation model that predicts in a single forward pass.
- Key idea: pre-train on synthetic datasets sampled from a prior over structural causal models; at prediction time the whole training table becomes context, approximating the Bayesian posterior predictive distribution without per-dataset training or hyperparameter search
- Topics: why tabular data resisted deep learning, scarcity of public tabular data, the path from AutoML and NAS to TabPFN, the TabArena benchmark, architecture changes from v1 to v3, scaling to larger tables, test-time compute, Google's TabFM, causal inference and interventions, relational data
- Also covers using TabPFN with coding agents; ends with a recorded update on the TabPFN-3.5 release, with a full reference list
More from Models
- Same-Named Claude Opus 5 Behaved Like a Different Model Hours Apart in Side-by-Side Test — FishingCharming5604 · 2026-09-24
- A 10-year trend holds: small fine-tuned models on selective data still beat bigger general models — xeophon · 2026-09-24
- GPT-6 Luna shows vision regression vs GPT-5.6: extraction drops 81.79% to 66.67% — ducha_aiki · 2026-09-24
- Luna 6 private coding evals don't look great — Maasu · 2026-09-24
- Claude Opus 5.5 tops Artificial Analysis at 58; four new models add 11 Pareto frontier points — ArtificialAnlys · 2026-09-24
- METR says it used an undisclosed 'additional source' to understand Anthropic's AI R&D, buried in the Opus 5.5 system card — coherence · 2026-09-24