

If you use internet discussions as training data, you can expect to find all sorts of crazy biases. Completely unfiltered data should produce a chatbot that exaggerates many human traits which completely burying others.
For example, on Reddit and Lemmy, you’ll find lots of clever puns. On Mastodon, you’ll find all sorts of LGBT advocates or otherwise queer people. On Xitter, you’ll find all the racists and white supremacists. There are also old school forums that amplify things even further.
Eventually, there’s going to be a February when they don’t send a bill at all.