As Brandon Harrera says: "Guns are awesome. Idiots are not."
So the reason why social media is so stupid is because people are naturally attracted to extremes. And when supply is high, only the most extreme outliers stand out. You don't care about a tiktok video where a guy does a thing - you care about a tiktok video where a guy does an extreme thing.
Sure, you might watch a video of a guy who can immitate the voices of 20 different singers, but you can't do that yourself
And teens are extremely impressionable. They see someone get a couple million views with a video where they chug a bottle of benadryl, and they want to one-up it, chugging two. Of course, what they don't see is the guy throwing up the thing afterwards, and thenn they land in hospital
> Instead of driving away new users i welcome KOKUYO and thank him for providing and sharing its content with all of us. You are welcome. -- ／人◕ ‿‿ ◕人＼ 苦情処理係 14:44, 28 February 2012 (UTC)
These companies don't care about what people say on their platforms until people start to boycott them or advertisers drop them.
And the government sees nothing wrong with that and lets that vicious cycle continue.
> @Michaeldsuarez you are very good at telling people what to think and what to do. Maybe you should apply as dictator in some country? I am sure you would do a great job. --Paddy (talk) 15:31, 28 February 2012 (UTC)
Like,I don't mind if a neighbor is a bit loud at 11pm when I'm still awake, but if it's 2am and I'm trying to sleep and he starts playing the fucking among us drip remix, then I'll go berserker
About how prisoners in concentration camps had their names erased and replaced with "~", meaning "name unknown", so "~ has been adopted by far-right extremists, implying that the holocaust never happened"
Too bad it never reached critical mass. I'd have been proud as fuck if some lefty mag printed an article about ~ being a hate symbol
We might be crossing the line tbh. I mean I know we're only talking about what's already on Wikipedia, but I think the terminology used might be... too explicit?
It reminds me a bit of a fingerprinting scheme, where something is burned through with electricity, and the path the electricity takes is different every time
I don't know much about crypto, but I know damn sure that crypto is hard and that I am not qualified enough to speak about it in anything but broad strokes
I can't find the Crystal guy (I'll link it later after asking in The Side Channel or something), but right now I'm just loving the trip down memory lane: nothingisreal.com/mentifex_faq.html
> To further confuse matters, Murray also has a tendency to rename his theory frequently; it’s variously referred to as the Concept-Fiber Theory of Mind, the Fiber-Concept Theory of Mind, the AI4U Theory of Mind, the Mentifex Theory of Mind, the Standard Model of the Mind, Project Mentifex, the First Detailed Theory of Mind, and the Grand Unified Theory of Mind.
And it made me feel like "Yes, now I understand some of the rough underlying concepts of the field of AI safety" and not like "Yes, now I know all there is to know about AI safety"
Yeah. But people misunderstand it and think it means paperclip maximizer.
Which is just total nonsense. Any AI capable enough of taking over the world would hack itself first to rewire its reward function and get really high.
Or he illustrates a concept how an AI may seem compliant during testing, but stops acting compliant outside testing, if there is only a small distributional shift
@forest You can't prevent people from having wrong ideas in a 10 minute youtube video
And very few people would actually watch a 2 hour long "introductory course"
I've long since recognized that a lot of people...are just not very smart
And explaining a topic as complicated as AI safety will just not work
Real perverse instantiation: Incorrectly labeling black people as apes and causing an outcry. Fake perverse instantiation: Doctor Who-style happy bots killing people who frown.
@MechMK1 There was one example where an AI was trying to learn how to land an airplane. It ended up exploiting a bug in the physics program where a high-speed nosedive into the ground at max power causes the integer storing damage to overflow, causing the virtual airplane to survive.
Another one was being taught to use a robot arm to pick up a ball and place it in the middle of a table. Because the reward function was the inverse of the distance of the ball to the center of the table, it ended up learning to rip the table out of the ground and tip it over.
One of the core concepts Miles discusses in his videos is that AI has a tendency to do things we might not want it to do, because we didn't state our goal clearly enough
It's the same with training animals. You might train your dog to sit when you say so and you give it a treat, but you're actually training your dog to be annoying and run around so you have to tell them to sit more.
One of the examples he uses was "Imagine you had a robot and you told it to get you a cup of coffee. On the way there, he breaks your vase, then returns with a cup of coffee. The robot AI would be convinced it did a good job, but you're likely very dissatisfied with its performance. The reason is because the vase exists outside the AI's reward function. It doesn't know you care about the vase, because you never told it that you do."
