CodeWP automatically classifies user prompts and messages, determining the best course of action, based on it's understanding of your message in the context of the general classification.
We have 9 internal classifications, which fall into 4 categories:
Generate code -> Ex; make me a new form plugin that sends it's submissions to a custom Slack channel.
Edit code -> Ex; change this so I can set the Slack channel in a custom Admin Page in my dash.
Explain code -> Ex;
Chat -> Ex; tell me about the most recent version of WordPress and what's changed.
For new user accounts, we've noticed that there are some situations where classifications are incorrect, especially between Edit and Explanations. Sometimes you expect code to be edited, but instead your changes are explained in a message, and vice versa.
CodeWP's classifier is a custom AI model that is self-learning, meaning the more you use our platform and provide feedback, the better it gets for your specific user account.
There's a specific process to improve the behavior of classifications:
1) Go to conversation settings.
2) Turn off Automatic Classification.
3) Submit your message, and Manually Classify
Note, even if automatic classification is off the first action will always be Generate Code unless you've pasted in code from an external source.
Additionally, don't hesitate to leave feedback on specific messages using the Thumbs Down / Thumbs Up feature. If you negatively rank a message, there may also be the option to leave specific written feedback. This is incredibly helpful and is used when we train future versions of the classifier AI mode.
If you want guaranteed accuracy, you can simply keep the Manual Classification on, and always choose your desired action.
Based on general platform feedback, we're constantly fine tuning our Classifier AI Model, resulting in consistent improvements in general accuracy for all accounts, out of the box. According to user feedback, over 90% of classifications are accurate, and we'll be working over the next few months to deploy updates and achieve a targeted 99% accuracy rate for our automatic platform message classifications.