I Challenged AI to Create a Better YouTube Channel Than Me... Here's What Happened
Вставка
- Опубліковано 24 сер 2023
- To try everything Brilliant has to offer-free-for a full 30 days, visit brilliant.org/adumb/. The first 200 of you will get 20% off Brilliant’s annual premium subscription.
Code from my videos available at: github.com/sponsors/adumb-codes/
Twitter: / adumb_codes
In this video I created an AI automated UA-cam channel to compete against a human run UA-cam channel to see which channel could get more views and subscribers in a month.
AI Channel: @TheDailyQuotation
Human Channel: @Quotables_
This video was sponsored by Brilliant
#ai - Наука та технологія
To try everything Brilliant has to offer-free-for a full 30 days, visit brilliant.org/adumb/. The first 200 of you will get 20% off Brilliant’s annual premium subscription.
There's a certain irony that YouTude named you @Human-279
Sup!
Importantly, this kind of long-form high quality thoughtful analysis can't even be produced by most people much less an AI.
ive seen ai shorts with 1000x the views of this video though
@@fergalhennessy775 tis luck in the algorithm
@@fergalhennessy775They are talking about quality, not success
@@silviocupica2521 nobody goes to yt shorts for quality lil bro
@@fergalhennessy775 You agree with him and then call him lil bro? ok kid
Plot twist, it turns out that adumb himself is an extremely sophisticated ai.
Yes
dude getting a sponsor in just your fifth ever video is such a huge flex
Actually nuts that you managed to get nearly 500 subscribers in a month on an AI channel
@@youbetyourwrasse if you like the content then what's the issue?
People love junk food.
thats how much subs i got on my youtube channel that i put weekly effort into between ages 14 and 17. not that im jealous or anything..
@@thedistinguished5255 tbf, short form content is a much easier way to get subs than long form
Nuts 😋
Congrats on getting sponsored already, your projects are awesome
Dude, you your creations are really next level man!
I'm sure I've come across a few of your reels before on youtube 😂😂
It's a real fight to get the algorithm to stop suggesting ai generated garbage
On UA-cam they are called shorts 🤓☝️
Don't think I really understood just how accessible this has all come when you can automate entire channels with AI already. Really interesting video and great motivation for me to continue studying Python :D
Uhhh...you get the AI to run python. You don't write it yourself.
This was a very interesting idea and the execution was amazing. Good video!
nicely done, showing pros and cons of an automated channel with actual data
Great video. I really appreciate the approach and thoroughness
This is a real gem of a channel. I love seeing this content, and you really inspire me to work on some cool hacks like this too!
This channel is really cool, man! Looking forward to seeing what you do next!
this is extremely interesting! keep up the good work
Honestly, using a script for editing but doing actual voiceover and research would split the difference on what matters.
A really neat video idea!!! Love your channel! Keep going!
This was so good! Thanks for making this
Very, very insightful analysis of a “real life” experiment, comparing the performance of an LLM versus a human. My take is that currently LLMs excel on quantity with “good enough quality “, but LLMs do not seem capable of “excellent” or “brilliant” performance.
That conclusion looks intuitive to me: when I look at the output of any of the LLMs I’ve tested so far, I do get something that looks as the “average opinion/knowledge” of what I see on the internet using a search engine. Thinking about how LLMs operate, this is what I expect: the tokens with the highest probability are those that are found most often on the internet (this is also the cause for bias). Thus, documents published by exceptional people (let’s say, some genius) will receive a very low probability of being generated by the LLM.
Anyway, I really enjoyed a lot this video, and I surely be looking for your future videos.
LLM's (unintentionally) are an exceptional way of representing the collective unconscious of WEIRD countries (Westernized, Educated, Industrialized, Rich, Democratic) and GPT 4 manifests this intangible "average" pretty eloquently.
The problem with this lies in the famous quote by Schopenhauer: "All truth passes through three stages. First, it is ridiculed. Second, it is violently opposed. Third, it is accepted as being self-evident."
Which is to say that yes, until the gears of society shift to accept the currently exceptional as codified and self-evident, the exceptional will remain fringe and speculative - outside the radar of LLM's.
Although the same (speculative and fringe) attributes can be assigned to the unfalsifiable ramblings of a conspiracy theorist. The benefit of humans is that we can physically test hypothesis in the real world and receive actual feedback and experience, and this is what differentiates genius from actual insanity: truth - both in terms of accuracy, precision, and frequency.
The point here is that to excel in excellence or brilliance, an AI (or AGI, to be more precise) would need to:
1. Be able to interact with the outside world and receive appliable feedback and experience from those interactions.
2. Be able to transfer that feedback to different (or unique, if taken one step further) scenarios, in essence mimicking the executive function and higher-level cognition of humans. (GPT 4 can already do the former, to an extent.)
3. Push boundaries and consequently face doubt, mockery, ridicule, or even violent opposition from the conclusions it generates.
Speaking of speculation, since we're on the subject of hypothetical future AGI, I feel it helps to have a rough idea of the difference between consensus truth, empirical truth, and axiomatic truth. In my opinion fields like psychology, political science, ethics, economics, etc are too "fuzzy" for AI to have an opinion on, or to try to influence - at least until we have a complete mapping of the human brain and can understand more empirically how the brain works in its entirety. Until then its genius should remain confined to pushing the envelope on empirical and axiomatic truth.
5:13 pov gen alpha: yay
2:04 The first suggestion from GPT reads like it's telling people "Don't go towards death. Become Death for others." Is it suggesting people should kill others? LMAOO
3:10 “ass=subtitles.ass” 😂
Your content is top notch, keep going and growing my guy.
dude this was such interesting and high quality video, well done
Amazingly implemented.
Man flexing he can make a large yt channel from scratch, but chooses to stay with this format for his integrity, hats off
This guy rocks! Amazing video
What I got from this:
Using automation to speed up the editing process is more effective than letting an AI make the entire video
Bro your channel is really interesting, you have high quality informative content, and I know how hard and time consuming it is
This is like a mathematical proof for the dead internet theory. I think of it being like checkers. Yes, checkers is a solved game mathematically, but we still haven't mapped out all moves. Thus, it still requires plenty of skill to play and be good at it.
"ass=subtitles.ass" 💀???
.ass is a respected file format
a s s
only a fool would disrespect the ass.
…or something 😛.
Clipclamp or whatever it’s called is useful because it can make short videos using stock footage and it already has a voice and it adds music
Incredible video, deserves more attention
Great video. Horribly depressing.
We are clearly doomed
the opportunity to find something inspiring, and yet you go for depressing.
Only if people decide computer > human
@@gavine2363Obviously it has already been decided
ai replaced the prayer service reflection at my school. not even religion is safe from the ai apocalypse
this was very interesting, good video
Youre still underrated as hell
might be worth re-doing this experiment now with GPT4 and more sophisticated tools overall
Really facinating.
Also, i've got an idea. Maybe if you were to make some human made content for some time, you can reach the monetization threshold and continue automation from there.
You're the NileRed of programming lol
bro the voice is literally the exact same
Man i hope you make it because i love the content
Simply brilliant!
Seems like this vid couldn't hack it in the algo but I really enjoyed it.
BUT YOU COULD JUST REQUEST the quotable api so it generates a random quote each time.
Based on the diagram in the "complete automation" section, that's exactly what they did
Tuvok was a great choice for the voice over AI.
Very useful experiment!
Great video!
Came here after watching your Wikipedia video…I think I just lost the rest of the workday to catching up on your channel 😂
same
I can't find it, but I remember that UA-cam had something in the terms of service against automated content (or perhaps against monetizing it). It's possible I misremember, but even if I do I think UA-cam would be likely to add something like that to its TOS if purely automated content had the potential to go viral.
Prtobably them Simps working there have no way to detect it.
It would probably be good for everyone... except youtube. What's the difference from youtubes pov? They have content that generates views. Why would they care if it is human or AI generated?
@@asdads3948 Probably reducing bloat on the servers? The problem isn't that they get a couple of viral bot channels that make them money, the problem are the millions upon millions of bot channels that do nothing but take up space on their servers. Because lets be real here, if you can fully automate a channel people won't just have one channel. There will be bot farms with 100+ channels constantly uploading junk content daily that all belong to the same group of people, multiply that by many groups doing the same thing and the situation can quickly become unsustainable. Banning fully automated channels from receiving money won't stop all channels, as there probably would be loopholes but it will stop a lot of people from even attempting it.
how does this only have 8k views 😭😭😭 this deserves millions fr
What if you create 10+ accounts to play the lottery youtube game in the attempt one go viral? (each account would have a different video, but the same style)
The king has returned
you were so concerned with whether or not you could, you never stopped to consider whether or not you should
I think some kind of combination would rock though
You could run the experiment a second time with completely new channels but incorporating everything learned from the first run. After introducing the new style of quote of the day videos on the ai channel, it gained a significant boost in subscribers. Restarting the experiment, you could try to investigate how helpful this boost was in the long term.
Google's "I'm feeling curious" is a really nice fun fact generator
This is the beginning of the dead internet theory
... I swear to god I've seen the short you showed in this video. (Specifically the quote one..)
Wow, this is such an interesting experiment! I love how you've created two distinct channels to compare the success they would achieve. It's fascinating to see the combination of handmade videos and AI-generated content in action. Both approaches have their own unique qualities and advantages.
The handmade videos bring a personal touch and creativity that can truly resonate with viewers. The effort, time, and craftsmanship involved in creating them can be appreciated, and it adds an element of authenticity to the content. On the other hand, the AI-generated channel offers a different perspective, leveraging the power of technology to generate intriguing quotes and facts. It's amazing to witness how AI can contribute to content creation.
I'm excited to see the results of this experiment and discover which channel garners more success. It's a great opportunity to explore the preferences of the audience and understand the impact of different content creation methods. Kudos to you for pushing the boundaries and exploring new avenues in content creation. Keep up the fantastic work, and I'll be eagerly following both channels to see how they progress!
(Written by ChatGPT)
Awesome video, as a dev myself I was wondering if it's possible to create something like this. Hope you will make it on this chanel with more interesting projects!
tbh i don't like automated channels i usualy block them i am on UA-cam for the people behind the videos :)
If you wanted to make the AI part a bit smarter, train it to optimise both views and subs, and adjust its content accordingly
You've earned a sub
We need to all band together to agree on a protocol when we come across an AI video. Either instantly scroll past, or dislike and then scroll past.
I usually block the AI channel and then scroll past the video.
Amazing vid
do the youtube analytics tell you how many people click "don't recommend channel" it would be in interesting metric to compare for the human vs ai content if you can.
I hit "don't recommend channel" for a lot of automated content, since the content isn't good, and the AI readers still don't seem to get sentence structure properly. They can't emphasise words correctly in a sentence, which is particularly maddening when they try to read stories.
here's a better idea for this project : if you upload one good quality human made video once a week and let the ai make two videos everyday this will make people get subscribed to you from the quality video while watching the ai videos so ur channel doesn't become idle this way the youtube algorithm will get your videoes suggested much more.
The most amazing thing about this is that google didnt immediately delete the channel
You could vastly improve the ai videos just by using better ai music and ai voice just with the huge improvement in ai since this video came out
Good luck 🙏
Almost 13 hours and this video is still at 900 views. :(
Now, this seems as a cool channel!
Quoting myself: The ai just cant compete with the target audiance's content
I think that the algorithm is really luck based, especially since you ran it just a month.
Now it’s time to finally answer the question: «Should developers make mobile games? Probably not. Yes they make more money but the reason can be described in a single sentence…»
i feel like youtube should lock you to 1 video a day
Do you plan on doing the experiment with long form contents?
I wish you had given the actual amount of money you made from each channel... that is an important piece of the puzzle
It shouldn't be one or another. We should take the best of both worlds. We can come up with the creative and original ideas and use AI tooling to speed up the production process
This is awesome
A pretty interesting stuff. Your content is really awesome
You became my favorite youtube channel
I would have liked to know how much the AWS cost for that pipeline was. I'm guessing not that much.
I did the same thing, but I quickly realized it would cost quite a bit of money and probably never generate it back.
How much money did this experiment cost?
I think it would also be fair to give AI feedback in numbers of views so it can improve its videos somehow. I suppose it would be hard, but maybe then it would have some viral videos and thousands of views.
Monetising shorts is actually damn hard, Google is keeping so much money
The conclusions about quality by each method are unsurprising, though I'm inclined to think that the relative effort required far more than makes up for the difference, no? Your graphs at 11:38 seem to bear this out. If the goal was strictly to monetize from youtube directly, indeed this isn't a useful structure, but then monetizing from youtube directly is a pretty foolish goal IMO. I'd think building a regular audience that can then be leveraged elsewhere (merchandising? creating a consumer/email list for something else?) is a better way to go. Curious your (or anyone's) thoughts.
Plot twist this video was made by AI
you should redo this video at some point in the future to see whats changed as more AI tools become available and existing ones get better (or worse in some cases)
However, subs/views doe not necessarily mean "better" ... That's a super-loaded word and a juicy, cool, uplifting and informative channel is something very different from an addictive collage.
Crazy to think that soon with AI video, we won’t even need these steps…
were the Only Up clips royalty free, or did you take them from some random streamers? i guess if you really wanted to automate it you could also have a robot play the game
Which video editing tool do you use to make your UA-cam videos on this channel? :)
I think you should upload more with the AI channel, a more realistic test would be 100 videos per day as it is the maximum amount of video you can upload with a classic API key.
On youtube quantity > quality by FAR. You can consistently make 100.000 subs in a month by uploading a few thousands times a day the exact same video, so with AI it can be way better and avoid the ban hammer.
And if you want to make money, as you said monetization is really tough. Getting the channel certified and selling it is a way, but there are only so much buyers out there, and it takes a lot of time.
Few issues here that makes me want to see the experiment repeated. First of all, the AI channel says it's an AI channel, and particularly for something like "inspirational quotes", there is going to be a bias against AI. Then there is the description, Evil AI vs Friendly human? This language reinforces the bias. There should be nothing on either channel saying if they're human or AI, leave that up to the viewer to determine, or at the very least say friendly AI and only in the description, not the title. Then there is the use of ChatGPT. This is a highly censored corporate AI, because of this it's been scrubbed to the point where it can no longer generate imaginative or relatable content. This is a side effect of heavily censoring AI, they lose their ability to be fun, entertaining, and relatable among other things. If you check out a project called Neuro-sama, you'll see AI that can be highly competitive with humans in creating content when the AI is made to be entertaining instead of being an answering machine and corporate PR representative. Not suggesting you make your own AI, but Huggingface has tons of uncensored AI, and you'll instead have to make a filter that insures it won't upload anything that's against the YT TOS, which should be fairly straight forward, and probably wouldn't even be needed if it was tasked with inspirational quotes or fun facts. The point is, what OpenAI and Google and all those does to their AI when they censor them isn't just making sure it won't be a racist or whatever, it takes away a lot of other capabilities too, it's basically making it dumb, dry, and fake/robotic/generic sounding, along with taking away it's ability to do more advanced reasoning and creative thinking. Neuro-Sama consistently drops very creative and weird things, sure, some of it is nonsense, but some of it is also highly quotable. One of my favorites is "Euthanasia and murder, a distinction without a difference", works on several levels, it's brilliant, funny, and to the point. But a corporate AI could never say that because it might be controversial, so it'll just say "As an AI I don't have opinions, but I understand this a complex subject with varying opinions and there is much discussion around this topic blah blah blah". And if you ask it to write an inspirational quote about euthanasia you'll probably get a "I'm Sorry Dave, I'm afraid I can't do that".
You lost me at _"what if I told you",_ = which, of course, is the line that marks every *_salesman_* on this Planet, as bad as on Ferenginar.
Any time I hear _that_ specific segment of syllables, I automatically *know* the speaker is all about trying to scam me out of my money.😀 So, bad for you!! 😂😂
Plot twist: this is the ai channel
Putting this aside, the free version of Chat GPT seems to get stupider every day. It might be just me, but I don't think so because it makes mistakes on basic math questions like asking it to add up numbers 1, 2, 3, 4, 5... 10. To be exact, it does the calculation manually by writing down the equation in a regular format (which is understandable) but it makes mistakes in writing by duplicating the numbers and then gives the wrong answer. I don't remember Chat GPT making such weird mistakes just a few months ago.
Petition for him to continue the AI channel
I find this really interesting, and I question how AI videos will fare long term- I started disliking, skipping, and "do not recommend this channel"-ing AI videos a few months ago and basically never get them anymore. They all end up being the same thing repeated over and over and it gets really annoying really fast.
Anyways, you just got another subscriber.