I attempted to change the first review to get a negative response but was not successful. What went wrong?
Try rerunning cell 2 and the cells below, because youâve changed cell 2âs content.
I found itâs possible to fool the LLM. For example, âDeadliest soup dumplings in Newfoundlandâ is found to be negative, but âdeadlyâ is a positive sentiment - in Newfoundland. Itâs understandable that the LLM didnât know this, but is there a way to tune it so it gets better at discerning nuances?
Has anyone run into a similiar issue - I canât open Amazon Sagemaker studio, and this is the error I see:
User: arn:aws:sts::285884772427:assumed-role/voclabs/user2920912=Michael_Ranasinghe is not authorized to perform: sagemaker:ListDomains on resource: arn:aws:sagemaker:us-east-1:285884772427:domain/* because no VPC endpoint policy allows the sagemaker:ListDomains action?
You could work on better prompting like for example if you know about the location of where the question comes from, or anything about who is asking, you could optimize the prompt for that.
Another, and perhaps a more solid solution, would be to fine-tune a model with pre existing reviews with their correct sentiment. I would choose a model suitable for the task to make it more efficient.