metadata

title: Key Takeaways
description: The key takeaways from this exploration
date: 2022-01-26T00:00:00.000Z
thumbnail: images/huggingface_logo.png

Conclusion

Here are some of the main ideas we have conveyed in this exploration:

Defining hate speech is hard and changes depending on your context and goals.
Capturing a snapshot of what you've defined to be hate speech in a dataset is hard.
Models learn lots of different things based on the data it sees, and that can include things you didn't intend for them to learn.

Action items?