Zero Shot LLM
We've built a large language model that can interpret our detection signals to make an attack/safe judgment on a message.
What is the item?
- We've built a large language model that can interpret our detection signals to make an attack/safe judgment on a message
- This model reviews messages in a text format, which is more similar to how a human reviews a message than how a traditional ML system reviews a message
- This increases the resilience of our overall detection engine
Why is it helpful to our customers?
- This large language model enables our system to process messages more deeply and with broader context
- This results in a substantial reduction in false positives and increase in attacks detected across all of our customers
Why is it interesting?
- This kind of system performs classification in a very different way from a traditional machine learning pipeline
- This suggests a new paradigm for detection systems that combine traditional machine learning with generative AI