4 ms·
I'd think any natural language model would have the same biases we see from real humans.
by Gigachad 5y ago
I'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.