Content check moved to a current AI model, and a size limit on browser cache
Two code fixes that came out of the 1 October maintenance. The AI model behind the message content check had been retired by our provider, so every check ended in an error. We moved the check to a current model. Each linked number's browser cache now has a size limit, so it cannot fill the disk again.
Highlights
- The content check no longer calls an AI model that has been retired
- Browser cache for each linked number is capped at 100 MB
In this release2
- FixedTemplates
The content check no longer calls a retired AI model
When you write a template, we translate the text to English and run it through a content check before it can be saved. The model used for the translation step, chatgpt-4o-latest, has been retired by OpenAI. Every check sent to it came back with a model not found error, so the check never ran.
What we changed:
- The translation step now uses gpt-4o-mini, a current model that handles short translations well.
- Nothing else about the check changed: the same categories are flagged and you see the same warning in the editor.
The check is not back yet. It is waiting on a separate billing issue with our AI provider, and it will start working as soon as that is settled. We will update this note when it is.
BeforeEvery content check asked for a model that no longer exists and ended in an error.
AfterThe content check asks for a current model that the provider still serves.