Google TimesFM 3.0 tested: zero-shot forecasting, 25% lower error, and two versions give opposite answers
Prompt Engineering · youtube · 2026-10-22
- Google released TimesFM 3.0, a zero-shot foundation model for time series forecasting that needs no training on your data; open weights on Hugging Face (non-commercial license). It tokenizes series as patches — "autocomplete for charts."
- The author explains what changed from 2.5 to 3.0, then benchmarks both on real NYC subway ridership (MTA data) and a simulated business running on Neon database branches: 3.0 cuts error 25% with 0.1s inference.
- When the forecasts are handed to a Gemini agent for a planning decision, versions 2.5 and 3.0 produce opposite answers. The video also covers 4 practical limitations.
More from Models
- ChatGPT web can actually listen to audio, but the same model can't via API or Codex — Possible_Statement84 · 2026-10-23
- MiniMax M3.1 Flash free for 3 days on 0G, with 1M context and 512K output — MiniMax_AI · 2026-10-23
- TLA+ expert corrects Grok: TLC only exhausts finite state spaces, not all behaviors — tianyin_xu · 2026-10-23
- Open LLM cost calculator shows Arabic tokenization and 6x price gaps across models — Moham-saed · 2026-10-23
- OpenAI blocks manual reset with 13% usage left: 'no reset needed right now' — AccomplishedSugar490 · 2026-10-23
- Meta's 816B Muse Spark Open scores 39, losing to 180B Qwen3.8-Flash-Next at 40 — BTA_Labs · 2026-10-23