I’ll be working on developing a plugin for Discourse this summer that will analyze a writer’s post before they submit as a means to reduce toxicity in a forum. I came across your most recent publication, the Unsupervised Sentiment Neuron. I started a discussion on the Discourse meta channel about the plugin idea and have received generally positive feedback/interest so far. I, along with other Discourse community members, think that the platform would make for a great training set for your team to work with! Google and the NYT are also currently experimenting with ways to automate moderation / lessen the burden of expensive human moderators with their perspective api. What do you guys think? Any thoughts?
Hi I work for Discourse.
Just wanted to chime in and say I think this could be a really fun project since it’s something every participant of discuss.openai.com could help dog-food every day, seeing as this is a Discourse forum.
We’d be happy to talk to any OpenAI researchers who’d be interested in working with us and @deevolutionism on this project. I’m sure many of our customers would be very interested in this project, e.g. Imgur and BoingBoing.