Google just dropped an open-source model that basically lets you predict anything.

@jakebeau_

19.8K followers · usually 1.6K views a post

Views

92.2K

Likes

4.6K

Comments

37

Shares

951

Selling · SaaS

TimesFM https://github.com/google-research/timesfm ↗

Why it worked

The video successfully bridges a complex technical topic (time-series forecasting) with relatable, everyday examples like sleep scores and productivity, making the product feel accessible and useful to a broad audience.

Summary

The creator introduces Google's open-source time-series forecasting model, TimesFM, and demonstrates how he uses it to predict his own productivity data. He explains the model's utility for various time-series applications and encourages viewers to share what they would predict using the tool.

Structure

1Introduce Google's new open-source prediction model
2Explain what TimesFM is and how it works
3Show real-world examples of time-series data
4Demonstrate personal use case with a Pomodoro timer
5Compare the model's predictive capability to Google Maps
6Conclude with a call to action for viewer engagement

Original caption

I’ve always said if I had a super power it would be to predict the future. Btw this is not encouraging speeding 🚨 Edit: version 2.5 was recently dropped, but earlier versions have been out for a while.

More from @jakebeau_

Their other posts in the index, biggest breakout first.

Product placement

TimesFM is the central subject of the video, shown through screen recordings of its GitHub repository and the creator's own data dashboard.

Call to action

So I'm curious, what would you predict? Drop a comment.

On-screen text

Google Sleep Score Open-Source Prediction Model Pre-trained TimesFM Decoder-only foundation model for time-series forecasting Predicts time series data I'm predicting my productivity Predicted focus time Trying to beat the prediction! AI is more than just LLMs Pre-trained ftw What would you predict?

Transcript

Google just dropped an open-source model that basically lets you predict anything. And here's how I'm using it. It's called TimesFM and it's a pre-trained model that lets you predict time series data. So think stock prices, your sleep score each night, or your weight over time. Since I've been building a Pomodoro timer, I use the model to predict how much I'm going to focus over the next seven days. I thought it would be cool to see the prediction and then try to beat it. Sort of like how when you're using Google Maps and you type in your destination, it tells you your estimated time and sometimes you try to beat that. AI is more than just LLMs. A lot of these other tools are getting overlooked, but I think these prediction tools are awesome, especially because you don't need a lot of user data to train them and actually make good predictions since they're pre-trained. So I'm curious, what would you predict? Drop a comment.