4 ms·
> OPT-175B has a high propensity to generate toxic language and reinforce harmful stereotypes So they trained it on Facebook comments?
by yosito 5y ago
> OPT-175B has a high propensity to generate toxic language and reinforce harmful stereotypes
So they trained it on Facebook comments?
- MengerSponge 5y agoDoes it merely reinforce harmful stereotypes? Or will it help perpetrate genocide?
- rhizome 5y agoTomato, tomahto.
- Gigachad 5y agoI'd think any natural language model would have the same biases we see from real humans.
- tsol 5y agoAre there really no moderated forums that the data can be taken from? Even HN-based training data would be much more civil
- Gigachad 5y agoA model trained on HN would spit out a 5 paragraph story about how minorities provide a negative ROI for cities. Or how the homeless need to removed from society.
- IAmEveryone 5y agoSure, but it would never do something actually bad, like raising the possibility that sexual harassment might, sometimes, be an issue, or questioning the value of phrenology.
- can16358p 5y agoDon't forget that it must also generate, at some point regardless of the topic, a new terminal emulator, and an extremely positive or extremely negative opinion about how blockchain can solve a problem.
- sanxiyn 5y agoNote that HN is included in the training data, see page 20.
- IAmEveryone 5y agoGo figure (8)!
- yosito 5y agoI'd think the training data is something that could be curated. Eliminating all bias might be impossible, but GIGO applies.
- Rebelgecko 5y agoI thought Pushshift was only reddit comments?
- stephenroller 5y agoWe trained on Reddit comments and HackerNews comments.