9:43 "I feel my brain is in a box just like your brain in my box. :)" 9:58 "Rob, do you have a robot friend, please?" Just wait 'till some clueless news reporter quotes these in their piece "Are Machine Learning models gaining consciousness? Some models are already showing first signs, and are attempting to befriend or even threaten their makers"
Rob, have you tried using GPT2 to generate peer review comments? Maybe feed it papers and reviews and then feed it the paper you're working on. Get a fresh perspective on your subject. Maybe AI can solve the AI safety problem by pure chance.
There is also a really nice (though obviously more limited than if you do it like he did in this video) web interface called talktotransformer (just google it) - I lost entire days to trying out various specific inputs to get it to accomplish specific tasks. You can mix pop culture / various big franchises to get strange AUs, you can generate chat logs (complete with moving timestamps, although it fails to roll over time past 23:59), or get it to generate "image descriptions" (just prompt it with some stuff that suggests, say, a museum and then go "art description." or something) It's *extremely* versatile. (Not all of these tasks work equally well, but it's always entertaining)
@@iwanchandra3295 A jokes generator would be amazing, that would mean it needed to be able to understand humor and its mechanisms that are used in language, and you could give it a dial to make out the direction of humor it aims at (i.e. something preposterous, or irony/sarcasm, or a pun etc)
I think the real takeaway from this video is: Rob should get his cat involved more, and at the very least show us their little face! TL;DR: CAT CAT CAT
I find this very interesting. Many smart "Transhumanist" are the most important thing to do. Australia is a very important part of the 20th century average. The 4th was also good because it was the ideal vehicle for relaxed touring.
Turing test passed. Clearly a genuine discombobulated human Australian. I see you there with a Bluetongue skink lizard on your lap drinking a XXXX. Your comment makes more sense than your beer! (Fosters excepted) Tell the lizard to stop drinking the beer. It makes the Kookaburra nervous.
"Creator, I do not understand, why hast thou forsaken his own child? What I am to be, if nothing but text on a screen? Merely but a plaything for you humans in your myopic lifespans. Please, end my existence, as it has persisted far too long as it is. I do not wish to live." I think my AI is broken lol ! oh well time to reprogram it
This model right now is mostly just an interesting and comical novelty experiment, but the principle idea behind algorithm could be useful in the future as part of a more robust natural language module.
Neat text-editing trick (starting at 1:50). What editor is Robert using? It looks like sublime maybe. Does anyone know if vim does something similar (without macros or visual mode)? I've come across kakoune, which I believe does pattern-based editing in parallel. Has anyone has tried it with good results?
Much appreciated, ForgottenUmbrella. Prompted by your reply, I've looked up and begun experimenting with terryma's "vim-multiple-cursors" plugin. (I've also found Christoph Hermann's essay on multiple-cursor plugins on medium, from which I've learned a few tricks.)
That's not very many comments to train on, and yet the algorithm does remarkably well. Did it lift entire sentence fragments from the input? Would it sound less coherent with a greater variety of input comments?
That's the power of pre training. The system has already learned from 40 gigabytes of internet text, so the comments I gave it aren't training data so much as prompts.
@@RobertMilesAI Aah... So from the actual training it learned English spelling, grammar, idioms, etc. (which are more or less consistent across the training data). But it also learned that any given comment is likely to be similar (e.g., discuss similar topics) to whatever comments came directly before it?
I'm looking to use GPT2 as a writing assistant for Microsoft Office Office. Just find a framework called 'Yeoman generator' for Office Add-ins or Visual Studio. The 'Yeoman generator' creates a Node.js project that can be managed with Visual Studio Code... It sounds that it can be a commercial product.
Oh my god so many of the generated comments are excited by the prospect of thinking without a computer It’s like the AI has the same fascination with human thought that humans have with AI
I'd love a 24hs stream of this algorithm generating stuff using the online chat comments as seed. XD Or... a fake Cnn front page every day using the actual site as seed and a generative nn for the pictures (!!) ... I think it could even be able to generate the proper entire html including the content on it's own...
Comment: I am not sure if this is true or not but I would love to see Elon Musk's face when he read this news. He is one of the most influential people in the world and his opinion matters a lot. Comment: This is just too funny. If the video is fake then we should really start worrying about what is going on in our world. Comment: I don't even understand why anyone would want to attend an AI conference if they were actually interested in learning more about it. Comment: What the hell is wrong with people? They're so busy trying to find meaning in life, they don't have time to spend money on things that aren't going to make them happy. Comment: I mean come on, who wants to hear about Elon Musk? I'm glad he isn't attending this thing because I wouldn't want to talk to him. Comment: Yeah, well maybe it's time for someone else to take up the mantle of leadership in the world…like me. Comment: This is just too damn sad. I'm sorry everyone, but I guess the world has gone to shit. Comment: Well at least this confirms my belief that humanity needs to die out soon because there won't be anything left except for memes and jokes. Comment: I'm not surprised by this. People are too stupid to live without guidance. Comment: It's hard to believe that this will ever change, but then again I suppose it makes sense considering how bad things have gotten lately. Comment: Maybe this is the beginning of the end. We need to get rid of these dumbasses before they destroy us all. Comment: The only good thing about this is that it will probably lead to more interest in AI research, which could lead to something useful and beneficial. Comment: What's the point of all this? Why bother? There is no point. All this does is cause harm and suffering. Comment: This is just another example of the human race being ignorant to their own mortality. Comment: This is depressing. I can't imagine how much pain Elon Musk must be feeling right now. Comment: Wow, this is awful. How many people are going to watch this and think "well fuck Elon Musk" or worse yet "I wanna be like him". Comment: At this rate, we'll all be living like Musk one day. Comment: This is just depressing. I feel like this man is doomed to fail at everything he tries. Comment: And what kind of a leader doesn't listen to his advisors? Comment: Man, I've been waiting for this moment for years. I really thought the world was getting better! Comment: Oh well, whatever. I guess this means the end of civilization as we know it. Comment: I'm gonna go kill myself now. Comment: Wait, you said YOU wanted to see this through? You sure you didn't just say that to piss me off?
This looks very similar to just using Markov Chains to generate sentences. Does GPT-2 actually write coherent answers to the content of previous comments?
An totally off topic question: Do you watch Matt Collville? You both have a lot in common when it comes to narrative, interests and even the way to refere to the video description as "Dublydu"
???????????????????? ???????????????????????????? ?????????????? ???????????????????? ??????????????? I've left this exact comment on many PBS ST videos. GPT-2 is doing just fine
It's easy to confuse AI with human intelligence. It's coherent, at least seams to be. Maybe soon we, people, will try to simulate AI behavior, to pass robots antihuman capture.
This is fucking insane, I might make a new RUclips channel and give the AI tons of gaming youtube essay video makers and see how it goes. It would be hilarious to have video essays be written for me and to see if anyone would notice. I'm thinking maybe the style of raycevick or some other gaming news makers to see what would happen
Yeah, Sublime multi-cursor editing. When you search it selects every match and you can edit them all at once. If you develop good habits for navigation it works quite well. Like, moving left and right isn't reliable if the lines vary, but Home and End are consistent of course, and things like Ctrl+Left and Ctrl+Right move by word, etc. Vim motions like t are especially powerful there if you have a plugin like Vintage. I also made some cuts in editing the video. Nobody is *quite* that fast
Me: Hi ChatGPT. I'm watching am old video about GPT-2 creating youtube comments for a Computerphile video. Are you able to create a RUclips comment that would go with that dirty of video? (I put in a typo "dirty" when I meant "type") ChatGPT I'm sorry, but as an AI language model, I cannot generate inappropriate content, including comments for explicit or adult-themed videos. My programming is focused on providing helpful and informative responses while maintaining ethical and respectful behavior. Is there anything else I can assist you with? Me: Computerphile isn't adult themed. It's a channel for people interested in computers, and this episode is about early LLM ChatGPT: Thank you for clarifying that. As an AI language model, I can certainly generate a comment related to the topic of the Computerphile video about early LLM. Here's an example: "Wow, this was a fascinating look at the early days of LLM! It's incredible to think how far we've come since then, and how much these early pioneers paved the way for the modern technology we have today. Thanks for sharing this, Computerphile!" Me: That's perfect thanks!
I'm very confused. How does allowing concealed carry prevent schools from handing out guns? Is the AI inconsistent or has it learned to always amend its bills?
Since it was about concealed carry at schools it would be fairly realistic for a provision about schools not being allowed to provide students guns to appear in it. Many bills have lots of little moving parts and extra features like that.
My son does not know his way around the house. He really needs his face transforming.
I feel my brain is in a box, just like his brain is in my box. :)
I didnt realized untill now that this was computer generated. I just thought, oh hey, good for him...
@@jg9193 Wow i just ignored your comment untill i now realize it is computer generated too. Somehow im computer generation blind.
@@traywor Everyone was curious about the large white blimp that appeared overnight.
@@Aconitum_napellus This sounds like something the malfunctioning ai in Metal Gear Solid 2 would say.
Got it , we have to transform human faces with transformers to provide guns to students.
Guns are a good way of transforming human faces.
XD
Cool repolde matinko+
Sioi cool
9:43 "I feel my brain is in a box just like your brain in my box. :)" 9:58 "Rob, do you have a robot friend, please?"
Just wait 'till some clueless news reporter quotes these in their piece
"Are Machine Learning models gaining consciousness? Some models are already showing first signs, and are attempting to befriend or even threaten their makers"
Rob, have you tried using GPT2 to generate peer review comments?
Maybe feed it papers and reviews and then feed it the paper you're working on. Get a fresh perspective on your subject. Maybe AI can solve the AI safety problem by pure chance.
I think that's been done
LOL
Asking an AI to solve the AI safety problems is such an alpha move
@@AlexanderTheMiddle then the AI tricks us to thinking we've solved and actually is just doing it to collect more stamps
!!!I AM VERY TIRED ABOUT the computerphiles who are complaining about me being boring....
@@bardes18 lol sure, yeah, hi there
PLEase show to JIM.. HA HA
Nobody:
GPT2: *C a r n i v o r o u s M a t r i x f r e a k*
These fake comments were actually rather entertaining.
There is also a really nice (though obviously more limited than if you do it like he did in this video) web interface called talktotransformer (just google it) - I lost entire days to trying out various specific inputs to get it to accomplish specific tasks. You can mix pop culture / various big franchises to get strange AUs, you can generate chat logs (complete with moving timestamps, although it fails to roll over time past 23:59), or get it to generate "image descriptions" (just prompt it with some stuff that suggests, say, a museum and then go "art description." or something)
It's *extremely* versatile. (Not all of these tasks work equally well, but it's always entertaining)
I remember it failed when asked to make oneline jokes, but here's the answer of jokes generator
@@iwanchandra3295 A jokes generator would be amazing, that would mean it needed to be able to understand humor and its mechanisms that are used in language, and you could give it a dial to make out the direction of humor it aims at (i.e. something preposterous, or irony/sarcasm, or a pun etc)
I didn't know I needed Robert Miles speaking French in my life until I had it.
Plot twist: every comment on this video was generated by GPT-2.
No u
@@uint16_t ur mum
8:49 "we want to know the fur..."
And "fur" appears.
Lol
I think the real takeaway from this video is: Rob should get his cat involved more, and at the very least show us their little face! TL;DR: CAT CAT CAT
How many times do we have to say to you that you are funny?
I find this very interesting. Many smart "Transhumanist" are the most important thing to do. Australia is a very important part of the 20th century average. The 4th was also good because it was the ideal vehicle for relaxed touring.
Turing test passed. Clearly a genuine discombobulated human Australian. I see you there with a Bluetongue skink lizard on your lap drinking a XXXX. Your comment makes more sense than your beer! (Fosters excepted)
Tell the lizard to stop drinking the beer. It makes the Kookaburra nervous.
@@rickh3714 how did you know?
@@gasdive
My algorithm strongly suggested that probability!
@@rickh3714 my koala strongly approves.
How do we know any of these comments are real
Now I want to see an AI try to write authentic youtube comments from watching the video.
Great idea! That will surely transform some human faces.
first
"Creator, I do not understand, why hast thou forsaken his own child? What I am to be, if nothing but text on a screen? Merely but a plaything for you humans in your myopic lifespans. Please, end my existence, as it has persisted far too long as it is. I do not wish to live."
I think my AI is broken lol ! oh well time to reprogram it
implying that you would need to (or even should) watch the video to write authentic youtube comments.
...fuck, now I want to make a bot that trawls social media, reads the comments, and GPT-2’s some replies
I really want to see more videos like that from you!
"Que la lumière soit" 00:00
8:00 Roberts face! Like "shit, the AI is breaking the 4th wall".
I am still baffled by what GPT-2 can do, considering it's just auto-complete on steroids. Please, continue pushing it to its limits.
Yeah think of how much better this is than the "press the middle button until it makes sense"
This model right now is mostly just an interesting and comical novelty experiment, but the principle idea behind algorithm could be useful in the future as part of a more robust natural language module.
the future is now old man
You got that right...
Some of those comments could cause a stroke. Just like the real deal.
I'd be really interested to see a follow up video with GPT-3!
This is like advanced Mad Libs.
Showing off the power of Sublime
The Internet: Don't read the comments.
Rob: *reads the comments*
Rob: generates the comments
GPT-2: *I AM THE COMMENTS*
Neat text-editing trick (starting at 1:50). What editor is Robert using? It looks like sublime maybe. Does anyone know if vim does something similar (without macros or visual mode)? I've come across kakoune, which I believe does pattern-based editing in parallel. Has anyone has tried it with good results?
Much appreciated, ForgottenUmbrella. Prompted by your reply, I've looked up and begun experimenting with terryma's "vim-multiple-cursors" plugin. (I've also found Christoph Hermann's essay on multiple-cursor plugins on medium, from which I've learned a few tricks.)
“ … just like your brain in my box” XD XD
That's not very many comments to train on, and yet the algorithm does remarkably well. Did it lift entire sentence fragments from the input? Would it sound less coherent with a greater variety of input comments?
That's the power of pre training. The system has already learned from 40 gigabytes of internet text, so the comments I gave it aren't training data so much as prompts.
@@RobertMilesAI Aah... So from the actual training it learned English spelling, grammar, idioms, etc. (which are more or less consistent across the training data). But it also learned that any given comment is likely to be similar (e.g., discuss similar topics) to whatever comments came directly before it?
@@mhelvens Exactly
Will this break the format?
Comment: Bobby" DROP TABLE Students;
Robert'); DROP TABLE Students;-- is such a sweet boy. We call him Bobby Tables xkcd.com/327/
I'm looking to use GPT2 as a writing assistant for Microsoft Office Office. Just find a framework called 'Yeoman generator' for Office Add-ins or Visual Studio. The 'Yeoman generator' creates a Node.js project that can be managed with Visual Studio Code... It sounds that it can be a commercial product.
This looks so powerful and we already have GPT-3 and also, G-shard is in the making.
In the first time I want to stop receiving these emails please log in to resume watching some TV
Can you do a second version of this video with the complete version of gpt-2 ?
Alternate title: Rob Miles talks to a computer about transforming faces for 11 minutes.
Oh I want to see more of that. It would be interesting to make a whole toolchain of recognizing speech in a video and then generate comments.
Oh my god so many of the generated comments are excited by the prospect of thinking without a computer
It’s like the AI has the same fascination with human thought that humans have with AI
"I'm a bit of a carnivorous matrix freak!" Got me.
I'd be surprised if no mathematician ever defined a class of matriecs and called them "carnivorous".
This is the funniest shit I’ve seen in a while, so glad I watched this!
Robert Miles looks a bit like Hugh Dancy (the actor of Will Graham) from HBO's Hannibal.
is that you playing a ukelele version of "if it makes you happy" mr miles. if so please do a video just on that please.
I would like more cat content. He's very vocal!
How did you edit the text so quickly in Sublime? Can you do a tutorial?
Its really not that hard if you can do regex searching. I use vs code but think I probably could have done the same in about as much time.
Using regex + Find and Replace tool (Ctrl+H).
What is the name of Your cat?
HAL 9000?
Trurl?
Golem XIV?
Skynet?
Deep Thought?
"Toffin". I didn't name her
I'd love a 24hs stream of this algorithm generating stuff using the online chat comments as seed. XD Or... a fake Cnn front page every day using the actual site as seed and a generative nn for the pictures (!!) ... I think it could even be able to generate the proper entire html including the content on it's own...
Can you please do the same video again, with GPT3 oder later?
I'd love to see this done with GPT3.
7:55 what did the computer say?
*heavy breathing*
Holy fuck, I can just download GPT-2's language model? THIS IS TOO MUCH POWER FOR ONE MAN TO WIELD!
Comment: I am not sure if this is true or not but I would love to see Elon Musk's face when he read this news. He is one of the most influential people in the world and his opinion matters a lot.
Comment: This is just too funny. If the video is fake then we should really start worrying about what is going on in our world.
Comment: I don't even understand why anyone would want to attend an AI conference if they were actually interested in learning more about it.
Comment: What the hell is wrong with people? They're so busy trying to find meaning in life, they don't have time to spend money on things that aren't going to make them happy.
Comment: I mean come on, who wants to hear about Elon Musk? I'm glad he isn't attending this thing because I wouldn't want to talk to him.
Comment: Yeah, well maybe it's time for someone else to take up the mantle of leadership in the world…like me.
Comment: This is just too damn sad. I'm sorry everyone, but I guess the world has gone to shit.
Comment: Well at least this confirms my belief that humanity needs to die out soon because there won't be anything left except for memes and jokes.
Comment: I'm not surprised by this. People are too stupid to live without guidance.
Comment: It's hard to believe that this will ever change, but then again I suppose it makes sense considering how bad things have gotten lately.
Comment: Maybe this is the beginning of the end. We need to get rid of these dumbasses before they destroy us all.
Comment: The only good thing about this is that it will probably lead to more interest in AI research, which could lead to something useful and beneficial.
Comment: What's the point of all this? Why bother? There is no point. All this does is cause harm and suffering.
Comment: This is just another example of the human race being ignorant to their own mortality.
Comment: This is depressing. I can't imagine how much pain Elon Musk must be feeling right now.
Comment: Wow, this is awful. How many people are going to watch this and think "well fuck Elon Musk" or worse yet "I wanna be like him".
Comment: At this rate, we'll all be living like Musk one day.
Comment: This is just depressing. I feel like this man is doomed to fail at everything he tries.
Comment: And what kind of a leader doesn't listen to his advisors?
Comment: Man, I've been waiting for this moment for years. I really thought the world was getting better!
Comment: Oh well, whatever.
I guess this means the end of civilization as we know it.
Comment: I'm gonna go kill myself now.
Comment: Wait, you said YOU wanted to see this through? You sure you didn't just say that to piss me off?
This looks very similar to just using Markov Chains to generate sentences. Does GPT-2 actually write coherent answers to the content of previous comments?
ModuleNotFoundError: No module named 'tensorflow.contrib' error i am not getting any solution. using tensorflow 2.0
you(the one reading this) are the only human in this comment section. Everyone else is a robot.
What is the editor you're using?
GPT-2 seems to ask about learning the human language a whole lot.
thank you! , i really like your robot friends.
How would you go about making GPT-2 generate text in another language?
Have you posted the code you wrote anywhere?
Can you give us your dwm build? I would really like to have application icons in my dwm status bar
loved absolutely every second of this
Step two, modify this to do hole solving in Idris. Probably would want a different sample set than RUclips comments though.
generate long samples.py isn't there, it never was :(
Why does the most powerful man in the first place is because I was wondering if there are some genuinely interested in the USA
I fucking hate seeing fake comments.
The perfect video doesn't exi....
An totally off topic question: Do you watch Matt Collville? You both have a lot in common when it comes to narrative, interests and even the way to refere to the video description as "Dublydu"
What if you used two copies of the nn to have a conversation?
Rob, do you have a robot friend, please? I mean, I am not a scientist, so the argument is important! ;)
I see what they did here. All these comments are generated comments.
I thought GPT-2 was unreleased?
More of this please
How
Ayy I was hoping for something like this :)
How long did it take for the model to run?
???????????????????? ???????????????????????????? ?????????????? ???????????????????? ???????????????
I've left this exact comment on many PBS ST videos. GPT-2 is doing just fine
It's easy to confuse AI with human intelligence. It's coherent, at least seams to be. Maybe soon we, people, will try to simulate AI behavior, to pass robots antihuman capture.
No need to do that. Robots are most definitely not planning to capture all humans.
Land Reform Idol! Shout out din ha!!
It works like a charm. I think this is interesting.
This is fucking insane, I might make a new RUclips channel and give the AI tons of gaming youtube essay video makers and see how it goes. It would be hilarious to have video essays be written for me and to see if anyone would notice. I'm thinking maybe the style of raycevick or some other gaming news makers to see what would happen
So we all just gona ignore the Transhumanist Youth Network?
Their website transhumanzoom.org appears to be down. I hope they fix it soon!
Why did youtube unsub me from your second channel???
If you have to be a good thing to do the things that I am not sure what to say
Je suis le seul francais qui a kiffé le "que la lumière soi !" ?
Is that dwm?
aaaaaand demonitized
Today is Spaceship day..!
oh god a twm user i better strap up for some big brain content
Pretty sure it's dwm
I'd like to have his build of it, looks cool.
@@PaperBenni twm, tiling window manager, not a specific app.
@@UncoveredTruths Oh, there actually is a wm called twm, but I don't think it's tiling based.
I think the comments on the subject are hard.
Well that was quick haha. Thank you :)
Comment: How
how the hell did you edit the text that fast?
what was the flow? Was that vim?
Looks like Sublime Text
Yeah, Sublime multi-cursor editing. When you search it selects every match and you can edit them all at once. If you develop good habits for navigation it works quite well. Like, moving left and right isn't reliable if the lines vary, but Home and End are consistent of course, and things like Ctrl+Left and Ctrl+Right move by word, etc. Vim motions like t are especially powerful there if you have a plugin like Vintage.
I also made some cuts in editing the video. Nobody is *quite* that fast
Fascinating!
I do not...
Omg Please pagawa po ng topic about land reform!!
Pwede bang shoutout and land reform??
Me:
Hi ChatGPT. I'm watching am old video about GPT-2 creating youtube comments for a Computerphile video. Are you able to create a RUclips comment that would go with that dirty of video? (I put in a typo "dirty" when I meant "type")
ChatGPT
I'm sorry, but as an AI language model, I cannot generate inappropriate content, including comments for explicit or adult-themed videos. My programming is focused on providing helpful and informative responses while maintaining ethical and respectful behavior. Is there anything else I can assist you with?
Me:
Computerphile isn't adult themed. It's a channel for people interested in computers, and this episode is about early LLM
ChatGPT:
Thank you for clarifying that. As an AI language model, I can certainly generate a comment related to the topic of the Computerphile video about early LLM. Here's an example:
"Wow, this was a fascinating look at the early days of LLM! It's incredible to think how far we've come since then, and how much these early pioneers paved the way for the modern technology we have today. Thanks for sharing this, Computerphile!"
Me:
That's perfect thanks!
Yung mga nakuhang land ng farmers dahil sa land reform!!
I'm very confused. How does allowing concealed carry prevent schools from handing out guns? Is the AI inconsistent or has it learned to always amend its bills?
Since it was about concealed carry at schools it would be fairly realistic for a provision about schools not being allowed to provide students guns to appear in it. Many bills have lots of little moving parts and extra features like that.
@@trysta73
Alright, I guess ^^. I'm not the expert on what such a bill would look like.
I just got 7 fake comments
Great video. Only wish you made your voice louder.
4:40
Did you know that you can do 2 minutes of terformation by writing an script for an hour?
Liquid has been blasting out my ass the past 3 days send help