TimeGPT: Machine Learning for Time Series, Made Accessible

Max Mergenthaler and Azul Garza Ramirez on TimeGPT, a frontier model for time series applications.

Subscribe: Apple • Spotify • Overcast • Google • AntennaPod • Podcast Addict • Amazon •  RSS.

Max Mergenthaler (CEO) and Azul Garza Ramirez (CTO) are co-founders of Nixtla, a startup that seeks to make cutting-edge predictive insights widely accessible. They build tools that  eliminate the necessity for organizations to maintain specialized teams of machine learning engineers, thereby streamlining the integration of advanced predictive technologies into various business operations. In this episode we discuss TimeGPT, Nixtla’s new frontier model for time series forecasting. TimeGPT can accurately predict future values for novel time series, using only historical data points. This promising “zero-shot” capability eliminates time-consuming retraining.

Subscribe to the Gradient Flow Newsletter

Interview highlights – key sections from the video version:

1. Target users for TimeGPT and how to use TimeGPT
2. Formatting data for TimeGPT
3. High-level overview of the training data for TimeGPT
4. Evaluation and benchmarking
5. Latency and inference speed
6. Domains and areas where TimeGPT has been successfully used
7. Suitable for forecasting and anomaly detection
8. Fine tuning TimeGPT
9. Proprietary nature and other early feedback
10. Potential impact of TimeGPT on the forecasting field
11. Timelines for TimeGPT-2 and the size of TimeGPT
12. Mixture of experts and retrieval augmented generation
13. More on the training data used for TimeGPT
14. Suitability for real-time applications
15. Automation and the future of time series analysis

Related content:

• A video version of this conversation is available on our YouTube channel.
• Federico Garza and Max Mergenthaler Canseco: A comprehensive suite of open source tools for time series modeling
• Philipp Moritz and Goku Mohandas: Navigating the Nuances of Retrieval Augmented Generation
• Ivy: Streamlining AI Model Deployment and Development
• Best Practices in Retrieval Augmented Generation
• OpenAI Developer Conference: Customizable AI Sparks Excitement and Concern
• Expanding access to Frontier Models with software and hardware optimizations
• Open Source Principles in Foundation Models
• Michele Catasta: Software Development with AI and LLMs

---

If you enjoyed this episode, please support our work by encouraging your friends and colleagues to subscribe to our newsletter: