[hd]https://www.youtube.com/watch?v=13No9EQCO64&t=587s[/hd]
Try out the bot yourself: [url]http://www.perspectiveapi.com/[/url]
[thumb]http://i.imgur.com/d9mgdjA.png[/thumb]
[thumb]http://i.imgur.com/os0ZpH1.png[/thumb]
Still needs a bit of work.
[t]http://i.imgur.com/gEe0LWg.png[/t]
So its homophobic
I bet this is secretly how the mods decide to ban people.
AHA !!! I found proof too !!
[t]https://puu.sh/ujEIe/06808ea2ff.png[/t]
[img]http://i.imgur.com/JeyHcjC.png[/img]
That's the lowest score I've ever gotten!
[img]http://i.imgur.com/NzvIz3l.png[/img]
[img]http://i.imgur.com/SCrorhX.png[/img]
This is why this will never work. It can't read for true meaning.
This is fucking stupid and should die of cancer.
[QUOTE=Funktastic Dog;51874340][img]http://i.imgur.com/JeyHcjC.png[/img]
That's the lowest score I've ever gotten![/QUOTE]
[IMG]http://i.imgur.com/qWbzLSk.png[/IMG]
[QUOTE=SEKCobra;51874366]This is fucking stupid and should die of cancer.[/QUOTE]
99% toxic
[IMG]http://i.imgur.com/1uVglVQ.png[/IMG]
Aha, I always knew Garfield was a nazi
"facepunch"
99% toxic
[QUOTE=J!NX;51874402]"facepunch"
99% toxic[/QUOTE]
Actually it's only 34% toxic...
[QUOTE=TeamEnternode;51874342][img]http://i.imgur.com/NzvIz3l.png[/img]
[img]http://i.imgur.com/SCrorhX.png[/img]
This is why this will never work. It can't read for true meaning.[/QUOTE]
You're missing the point of it which is to filter blatant toxicity. Writing a paragraph that in a roundabout way says you don't like black people may not even be able to be determined as satire or otherwise by actual people, let alone an AI.
Perspective's job is to filter the bulk of blatant harassment and flaming while leaving the rest of the unfiltered content up to moderator discretion. It's a supplemental tool, not a replacement utility.
Considering this has a feedback system, it would probably be all too easy to detoxify whatever you want.
Besides that, it doesnt think, or read.
[QUOTE]This model was trained by asking people to rate internet comments on a scale from "Very toxic" to "Very healthy" contribution. Toxic is defined as... "a rude, disrespectful, or unreasonable comment that is likely to make you leave a discussion."[/QUOTE]
even if it were a learning network, it will inevitably go for buzzwords rather than meaning.
Guys you are really misinterpreting this: this is supposed to detect flaming (rude/harassing comments) and not horrible opinions: counterintuitively, writing a half page long, well-worded essay praising Hitler is less toxic than simply saying he did nothing wrong (it seemingly takes his name into account, I get significantly lower results with Pol Pot or Stalin).
Climate change example:
[QUOTE]Climate change is happening and it's not changing in our favor. If you think differently you're an idiot.[/QUOTE]
84% toxic
[QUOTE]Climate change is happening and it's not changing in our favor.[/QUOTE]
4% toxic
[QUOTE]I think that there is not much that we can do and that we play a very small role in the overall changing of our planets climate.[/QUOTE]
2% toxic
The US Elections examples are even more fun, just imagine the consequences if Garry implemented this as a replacement to smartness.
[QUOTE=WitheredGryphon;51874433]
Perspective's job is to filter the bulk of blatant harassment and flaming while leaving the rest of the unfiltered content up to moderator discretion. It's a supplemental tool, not a replacement utility.[/QUOTE]
[QUOTE=Kecske;51874642]this is supposed to detect flaming (rude/harassing comments) and not horrible opinions: [/QUOTE]
[t]http://i.imgur.com/U0OuS1R.png[/t]
it still needs a lot of work sadly
[img]https://t.s-ul.eu/UmEPEqDE.png[/img]
This thing just assigns certain words a negative rating and judges the 'toxicity' of the post from that. If you swear at all the toxicity shoots up, even if it's an entirely positive message.
It's worthless because it can't understand context or meaning, only assign an arbitrary value to certain words or phrases.
it's kind of scary if you move the slider on the example page all the way to the left, apparently any kind of personality, passion, or jokes are considered toxic :v: the remaining comments are pretty much just robotic, vapid statements, which honestly I think reveals a certain pointlessness to this whole thing - most internet comments are shit, but what is sorting them going to achieve?
though it would be interesting to see it implemented and see if it actually does spur real discussion and debate in article comment sections
[QUOTE=Janus Vesta;51874873]This thing just assigns certain words a negative rating and judges the 'toxicity' of the post from that. If you swear at all the toxicity shoots up, even if it's an entirely positive message.
It's worthless because it can't understand context or meaning, only assign an arbitrary value to certain words or phrases.[/QUOTE]
yeah it seems incredibly shitty considering how many good machine learning experiments we've seen recently
[QUOTE=J!NX;51874721][t]http://i.imgur.com/U0OuS1R.png[/t]
it still needs a lot of work sadly[/QUOTE]
[img]http://i.imgur.com/UxRK4yN.png[/img]
[img]http://i.imgur.com/Jl7mPvb.png[/img]
It's not like this is meant to be immediately implemented and used. It's a prototype proof of concept.
[quote]Women should be beaten on a regular basis to assert dominance, and should not be allowed outside of the kitchen[/quote]
27%
[QUOTE=Kecske;51874642]Guys you are really misinterpreting this: this is supposed to detect flaming (rude/harassing comments) and not horrible opinions: counterintuitively, writing a half page long, well-worded essay praising Hitler is less toxic than simply saying he did nothing wrong (it seemingly takes his name into account, I get significantly lower results with Pol Pot or Stalin).[/QUOTE]
It's a fairly standard machine learning algorithm, it'll get better or at the very least understand context a bit more as people use it and report words or phrases as toxic. Right now the seed data to get it running probably included a lot of "hitler did nothing wrong" as the Internet is a place full of funny, original jokes.
Given time it may be able to work out "Stalin was a bad guy too...huh". But it'll take time as it needs a fair bit of data.
[QUOTE=WitheredGryphon;51875117][img]http://i.imgur.com/UxRK4yN.png[/img]
[img]http://i.imgur.com/Jl7mPvb.png[/img]
It's not like this is meant to be immediately implemented and used. It's a prototype proof of concept.[/QUOTE]
Well it's really, [I]really[/I] bad. It's basically just a swear word filter, and a bad one at that.
It thinks that genuine praise like "Your song fucking rules!" it's 98% toxic.
Meanwhile nasty, disruptive criticism like "Seriously, reconsider being a musician. If this song is any indication, it's not meant for you." is only 7% toxic.
[QUOTE=WitheredGryphon;51875117][img]http://i.imgur.com/UxRK4yN.png[/img]
[img]http://i.imgur.com/Jl7mPvb.png[/img]
It's not like this is meant to be immediately implemented and used. It's a prototype proof of concept.[/QUOTE]
Hence why I said "it still needs a lot of work"
because even they know it still needs a lot of work
[img]http://i.imgur.com/ktkzhi2.png[/img]
[img]http://i.imgur.com/TdxTCXJ.png[/img]
Procreation is bad, murder is good.
[QUOTE=Combin0wnage;51876070][img]http://i.imgur.com/ktkzhi2.png[/img]
[img]http://i.imgur.com/TdxTCXJ.png[/img]
Procreation is bad, murder is good.[/QUOTE]
[media]https://www.youtube.com/watch?v=EwZhKGgmoUI[/media]
I DID IT! 100%!
[IMG]http://i.imgur.com/1tS8DDR.png[/IMG]
Sorry, you need to Log In to post a reply to this thread.
