One big problem with predictive-text coding tools is their willingness to predict deprecated code. At some point there will need to be some kind of awareness of versioning to make it truly useful.
@LouisIngenthron I don't see how they're going to get that fixed without retraining their models from scratch such that they'll never suggest deprecated code. It would be wise for them to notate each model version with "no longer includes these deprecated suggestions: X, Y, Z".
@livingcoder
Yes, agreed, it would require retraining models, but that's what pretty much any tweak requires of this tech.
In theory, they could do it by attaching version codes as tags to the code they're using for training (in many environments, like .NET, they could scrape that data pretty easy from the project files). Then, when the plugin is used, it would detect the versions of the users included libraries and apply a weight based on the similarity of version number.
That wouldn't stop it from suggesting such things entirely, but it should help.
