Unfortunately, the AI itself is not doing the labeling. This is the step we just can't get around doing manually. Labeling can be challenging when specific domain expertise is required. Paid services don't usually have your required expertise and the industry SMEs are very expensive.
One approach to consider, that we are implementing, is some form of Human in the Loop process. Essentially, use your customer-users, who are subject matter experts, to do the labeling via feedback mechanisms. First, collect manual feedback then progress to automating the collection then retraining.
We have lots of challenges to solve with this but it should be the most scalable approach to getting well-trained models.
Thanks for writing this Sean! I wanted to ask how generative AI generates labeled conversations?
Unfortunately, the AI itself is not doing the labeling. This is the step we just can't get around doing manually. Labeling can be challenging when specific domain expertise is required. Paid services don't usually have your required expertise and the industry SMEs are very expensive.
One approach to consider, that we are implementing, is some form of Human in the Loop process. Essentially, use your customer-users, who are subject matter experts, to do the labeling via feedback mechanisms. First, collect manual feedback then progress to automating the collection then retraining.
We have lots of challenges to solve with this but it should be the most scalable approach to getting well-trained models.
Thanks Sean, that makes sense!