3 ms·
Charlie Gerard made a Github Action recently that uses tensorflow's toxicity classifier to block some of the more horrible PRs and comments on repos - https://g
by onion2k 6y ago
Charlie Gerard made a Github Action recently that uses tensorflow's toxicity classifier to block some of the more horrible PRs and comments on repos - https://github.com/charliegerard/safe-space https://github.com/charliegerard/safe-space - it'd be good if Github could roll something like that out across the entire platform.
- Qub3d 6y agoWhile neat, this seems like an even more protracted version of the Scunthorpe problem just waiting to happen. We already know mass-automation of morals is problematic (see YouTube, Twitter, Facebook et al's attempts). This stuff is deplorable, but for now it requires human eyeballs and judgement. Perhaps allow for more granularity on repo visibility and interactivity beyond public/private.
- onion2k 6y agoWe already know mass-automation of morals is problematic (see YouTube, Twitter, Facebook et al's attempts). This is different. On Facebook, Twitter etc is that they're trying to use AI to automate detecting general speech. That's a hard problem. PRs and comments on GitHub aren't really general speech, which simplifies the issue, and GitHub already has access to billions of legitimate PRs to use as a dataset for training. It is at least worth exploring. I'll also add that while I think such an automated system should be opt-out by default for public repos there absolutely should be an opt-out if people want to use one.