Journals - thedailyminijournal.netlify.app/ or goalsontrack.com/- I believe the some long form journal is helpful for reseting direction, but this will not help me everyday. Source: about 3 years ago

Even the developer blog over at PerspectiveAPI link seems pretty light on details beyond things like flags for obscenity and sexually explicit content. Source: 6 months ago

Here it is, with a big ‘HOW IT WORKS’ on the front page if you’re into that sort of thing. Source: 6 months ago

To get to the heart of what sets tankies apart, we embarked on an in-depth comparison with other far-left online communities (depicted with a light red in Fig. 2 and 3). We used a range of tools to analyse the topics they discuss and the names they mention, and study their lexical choices by examining the word alignments of tankies with other far-left communities. Subsequently, we assess the toxicity of their... Source: 8 months ago

Prior to shipping our first LLM, we trained two smaller models tested offline for this exact use case. The first was a Logistic Regression model which performed relatively well on a training set containing ~120k labels. The second model was a Gradient Boosted Tree (GBT) model which outperformed the Logistic Regression model on the same training set. The tradeoff was speed both in training and inference time as the... Source: 9 months ago

Disqus is a large public engagement platform which hosts millions of comments per month. They’ve been developing their own technologies to better handle problematic content and help their users moderate. They also leverage Jigsaw’s Perspective API, which paved the way for ModerateHateSpeech. Source: about 1 year ago