Google's TimesFM forecasts almost any time series with zero training

Comment "FORECAST" and I'll DM you the link. Google's TimesFM forecasts almost any time series with zero training. You give it your numbers. Sales, traffic, energy use, temperature. It tells you what comes next, with a likely range around it. It was first trained on 100 billion real data points, including Google Trends and Wikipedia pageviews. The newest version, 3.0, came out in August on over 1 trillion, real and synthetic. Version 3.0 can also take in outside factors like holidays, weather and prices. You have to supply those. You can install it with a coding agent. Or use it inside BigQuery, and from Connected Sheets if your data is in BigQuery. Now the honest part. It can't predict "anything". It forecasts numbers over time. The range runs from the 10th to the 90th percentile. Real results can land outside it. The 3.0 weights are non-commercial if you download them. Use 2.5, or BigQuery, for business work. Rakhul Charts and demo: Google Research and Google Workspace. Some charts in this video are my own drawings of the idea, not real results. #timesfm #forecasting #timeseries #aitools #bigquery