My SFW public DNS service is now accepting quite a lot of traffic. This is fine, but the model needs to be periodically retrained to keep it accurate. Donating GPU/TPU cycles (on Google Cloud or whatever) would be immensely useful for this. Thanks. https://fr.dnscrypt.info/sfw.html
The exact same model can be trained with a dataset that includes violence, hate, etc. I haven’t done that, but its performance would likely be similar.