"But what would be the point doing this in the browser? Consider the fact that many of the Chrome AI APIs help with writing, I think such a solution could be useful in helping flag content that may be considered spam by others. So for example, your web-based CMS system may let you craft email content, and being able to see in real-time if the content would be diverted to someone's spam folder could be incredibly useful."
"To build my spam checking tool, I relied on the following features: The Prompt API, which is the general purpose aspect of Chrome AI. A system prompt that guided the Prompt API to focus on the task of determining if the content would be considered spam. Structured output so I could properly display the results. Here's how I did this. The first two aspects are handled by the code that creates the session object:"
On-device spam detection can run inside the browser using Chrome's Prompt API, providing immediate feedback on whether written content will appear as spam. Real-time checks are useful for web-based CMS editors and email composition interfaces to help avoid content that might be diverted to recipients' spam folders. The implementation uses a language model session initialized with a system prompt that returns a true/false spam judgment plus a list of reasons. Structured output formats the model response for display. A session monitor listens for downloadprogress events and suppresses updates at 0 and 1 to avoid spurious messages while showing progress.
Read at Raymondcamden
Unable to calculate read time
Collection
[
|
...
]