Skip navigation
image of an owl Alamy

Meet Owl: The Large Language Model that Automates IT Tasks

The new AI model could revolutionize IT operations by automating processes such as troubleshooting.

This story originally appeared on AI Business.

Researchers have introduced Owl, a specialized large language model trained specifically for IT operations. Unlike general-purpose counterparts such as GPT-4 or Llama 2, Owl focuses on performing a range of IT tasks, from network security to data analysis.

The 13 billion-parameter model was trained on a specially curated dataset called "Owl-Instruct," which is made up of IT-related information such as system architectures and network operations.

Owl is designed to help automate and enhance IT workflows, including improving data analysis and troubleshooting processes.

Owl could be used to automate or augment routine tasks like log analysis monitoring, and even some aspects of customer support, which the researchers suggest could free up time for IT professionals to focus efforts on more complex tasks.

Owl utilizes a so-called "mixture-of-adapter strategy," which is designed to enhance the instruction-tuning performance.

Effectively, it is like using various add-ons to help the model specialize in a particular task — in this instance, the strategy is used to help the Owl model learn the task-specific and domain-specific representations from diverse inputs.

The team behind the model also built the Owl-Bench Benchmark to evaluate AI model performance on IT-related tasks. The newly built benchmark consists of 317 Q&A and 1,000 multiple choice questions related to IT tasks across different domains including log anomaly detection.

Testing the Owl model on its namesake benchmark, the researchers found that it outperformed more general models such as Meta's Llama 2 and Alibaba's Qwen-7b. The Owl model was found to have achieved state-of-the-art results on IT tasks like log parsing and anomaly detection.


Owl was built by researchers from Beihang University and Cloudwise Research.

Hide comments


  • Allowed HTML tags: <em> <strong> <blockquote> <br> <p>

Plain text

  • No HTML tags allowed.
  • Web page addresses and e-mail addresses turn into links automatically.
  • Lines and paragraphs break automatically.