Against Time-Series Foundation Models: ex-Stripe forecaster bets on agents, not bigger TSFMs
anshulkundaje · x · 2026-09-04
A forecaster with stints at the Federal Reserve, Amazon supply chain, and Stripe argues that time-series foundation models (TSFMs) are struggling to beat 50-year-old statistical methods—and that bigger TSFMs aren't the future. Drawing on years of hands-on forecasting work, he predicts the winning approach will be general agentic models searching over specific forecasting problems and then fitting something closer to structural time-series models, questioning whether TSFMs are even solving the right problem.
More from Research
- Kangwook Lee: agent harness improvement can happen inside rollouts once agents are capable enough — Kangwook_Lee · 2026-09-04
- The Bayes Bandit: A Mathematical Take on Curiosity in Reinforcement Learning — CatAstro_Piyush · 2026-09-04
- Andriy Burkov Releases The Dark Hundred-Page Language Models Book on Kindle — burkov · 2026-09-04
- Alibaba-NLP CORE: boosting compositional reasoning in MLLM embeddings via reranker distillation — Alibaba-NLP · 2026-09-04
- On-policy distillation improves for hundreds of steps from a single query — it's algorithm-starved, not data-starved — Thinking-Space · 2026-09-04
- WorldReward: a vision-language reward model for evaluating camera-conditioned world models — Yibin Wang · 2026-09-04