3 ms·
If that's the plot, then I think Scott Alexander did it better with Sort By Controversial, a short story about a team who uses machine learning to generate maxi
by SilasX 6y ago
If that's the plot, then I think Scott Alexander did it better with Sort By Controversial, a short story about a team who uses machine learning to generate maximally-controversial claims/articles, then find themselves ensnared by that very same inflammatoriness.
https://slatestarcodex.com/2018/10/30/sort-by-controversial/ https://slatestarcodex.com/2018/10/30/sort-by-controversial/
https://news.ycombinator.com/item?id=21190508 https://news.ycombinator.com/item?id=21190508
- FrozenVoid 6y agoA state of art network like GPT-3 fails to appeal to human emotions and sounds artificial, so as a wedge-issue(the actual term for thing like 'Shiri Scissors') generator it wouldn't produce anything on the scale of human trolls(who actually understand psychology), besides the reputation of new user posting auto-generated trolling content will not be taken as seriously. A real troll would construct a narrative that is believable and appealing to a wide audience, without any controversial content: the controversy would comes from implications and analysis of statements in detail, i.e. the 'obviously controversial' parts are bad trolling that would detract from the position advanced: An AI trained to produce controversial content wouldn't understand such subtlety and produce directly inflammatory content.
- SilasX 6y agoThat would just mean that “controversial” isn’t the best term for “whatever that thing is they’re optimizing”, not that that thing is fundamentally machine-unlearnable.
- FrozenVoid 6y agoThe thing is, the learned data comes from humans who treat offensive/controversial/trolling/infflamatory content as 'controversial' and will argue against obvious bait topics for hours.