4 ms·
Really not a fan of these types of technology, with the subtleties of language such as sarcasm and irony, and then you've got approved narratives and taboo subj
by colordrops 9y ago
Really not a fan of these types of technology, with the subtleties of language such as sarcasm and irony, and then you've got approved narratives and taboo subjects, and those times where the minority is right and is under attack by the mob.
I'd only support this tech as a filter for human moderators and not as an automated system.
- dr_zoidberg 9y agoTwo examples that support your reasoning: > Your a socialist snowflake! Is ranked as less toxic than: > Please put yourself in the shoes of women, minorities, and LGBT people Extracted from the "US Election" demo on the page. Edit: for the sake of clarity, when taken to the textbox below, snowflake gets 58% toxicity, and minorities gets 29%. Still, when using the slider, snowflake appears before, with a circle (which would correspond to an ok comment), and minorities much later on, with a square (which would be a dubious comment). Edit 2: testing on the textbox, commas and periods get you a lower score (less toxic), while exclamation marks, lack of punctuation, and "bad writing" (not just grammar, but a general bad style you could say) get you a higher score. Edit 3: the final gem I'll take for my professors at University: FAT is an old filesystem! gets 90% :P Secret -- caps seem to also rise the score. Actualy... fat 88%, fat_ 41%... now that's it, I got tired of this.
- majewsky 9y ago> "Your a socialist snowflake!" is ranked as less toxic Maybe it tripped over the wrong usage of "your"? :)
- moxious 9y agoMachine learning is getting a bit out of hand. not in the sense of AI and replacing humans (that's not happening) but in terms of hype and over-application. This API may represent an attempted technological solution to a social problem. Those have a very poor track record. As a filter for human mods it might be an interesting incremental improvement. But google has a need to do this at a scale where I can't see it staying there in terms of the concept of deployment.