newsfilter.io
Conference Presentation, Keynote, Product Demonstration

Keynote by Christopher Savoie, Zapata AI CEO | RAISE Summit 2024 | Paris

  • Zapata anticipates continuing the use of forward-looking statements as a public company and plans to sustain the application of tensor networks for generative AI training and inference, diverging from traditional neural networks.
  • The company predicts that vector databases and tensor-based architectures will deliver faster, more compact, and high-performance storage for weights and biases at scale.
  • Zapata intends to deliver real-time insights at the edge by compacting vast operational and time-series data, with specific plans to demonstrate predictive capabilities using live streaming sensor data.
  • The organization expects generative AI evolution to shift from single large language models to ensembles of smaller language and non-language models working in tandem to create intelligent agents for complex tasks.
  • Zapata plans to demonstrate the effectiveness of tensor-based models on limited hardware, specifically two A100 GPUs, without requiring cloud connectivity involving billions of GPUs.
  • Current operations utilize an ensemble of five independent models to predict on-track events like accidents and yellow flags in F1 and IndyCar races, with claims of predicting a yellow flag probability one lap prior to occurrence.
  • To predict live lap times, Zapata plans to maintain 30 ensembles of two underlying models for each car on the track.
  • The company expects to generate synthetic data for physical properties lacking physical sensors with an error rate of less than 1%.
  • Plans include updating models live within military-grade server environments designed to withstand extreme heat and vibration.
  • Zapata expects the mathematical principles demonstrated in racing applications to extend to biomanufacturing, advanced manufacturing operations, and trading strategy determination.
  • The organization predicts that its time-series real-time updates will be applied to forecasting telecom network disruptions.