it might be your brain tricking you into thinking youre listening to someone incredibly well known or whatever so its moer interesting to you...im here for it, have a generated travis scott teaching little kids how to do calculus.
A bigger question and very politely asking, is why did your comment get 923 Likes? I believe I wrote an analytic comment dissecting the legal issues of this AI video and I only received 4 Likes.
The body is real, only the face is AI. Although the video insinuates it being generated, the face is deepfaked onto a real person who’s talking. “Only” his face and voice has been replaced.
@@TheChannel1978 In the future, people will be able to create their own full length movies just with ai. The ai can create any famous actor at whatever age you choose them to be and put them in the movie. Of course you will be sued if you try to sell this movie for money if you are using the actors image without permission but if you are just creating a home movie for fun then I think its completely fine.
@@PaulPavloPabloyep and eventually there’ll probably be no new human actors coz everyone will be wanting to use the ai versions of past ones and TV shows will probably generate original AI ones. It’ll be back to the theatre to see actual new human actors
I feel like most ppl focus on the neural network architecture which really has little to do with what makes modern LLMs so impressive since NNs have been around forever; honestly this might be the best high-level explanation of modern LLMs that I've seen.
Really? Words go into matrices, then coefficients are applied, then they’re juxtaposed with other matrices and then answers are generated. … I knew LLMs are complex but this “explanation“ basically just confirmed that there’s lotsa numbers crunching and lotsa databases involved. Duh…
Jesus christ, this was wild. I forgot several times that this is not actually Ryan Gosling giving me an intro to LLMs. Video still has a way to go but voice is very close to perfect already. Gonna be some wild years coming up.
This is the future of education. Breaking such complex topics down in such a playful yet not diminishing way is exactly what would help my concentration levels stay up during the LLM classes at uni
@@howtofly-with-AI genuine question, is this grounds for litigation? I mean you're using his face without his permission, or is it protected under some educational law or something?
How so? Just because we can program AI to iteratively incorporate existing/past information into generating something new does not really say much about the fact that the human mind does the same, nor does it say anything about whether it can be considered the same mechanism. To say that these programs can disprove free will is like saying this video disproves Ryan Gosling is real.
@hairyott3rr you are thinking very literally. Not-Ken provides a metaphorical example that helps understand the illusion of free will. And like he's using life to explain LLMs, his example is tautological and very clever. In life, gene expression and neurochemical responses can be seen as tokens. Each event can be expressed (metaphorically) as tokens that form future responses and events. We can not redefine the tokens of the past, and therefore cannot reconstruct the future through free choice. It is a metaphor that is useful, not scientific proof, but an accessible way to illustrate a philosophical point.
@@shaunshelly3314 you keep insisting that free will is an illusion, but that's only a presupposition not a fact. it's more precise to say that this ai process shows an alternative to the idea of free will.
@foolfether these programs tell us nothing about free will. For me. It is a metaphor - and a fairly good one. The free will argument is largely philosophical. But since you seem intent in going there: 1. You have mo control over when, where, if you are born. 2. You have no choice about your genes or gene expression even after you are born. 3. You have mo choice over your personality, which is largely fixed. 4. You have an illusion of choice over many things. But from any perspective, the choices you may have are very limited. By environment, circumstances, and the sum of your past experiences. 5. If I put you in an fMRi machine and asked you to decide to touch a button in your left hand or your right. I could tell you 100% of the time what your 'choice' would be before you even decided. So, if we have any freewill, it is limited to the narrowest of bands. People can only make the choices they can make. To illustrate my opinion in a so that some people may better understand, I could use the LLM as a metaphor. [I could say that all our experiences and actions can be broken into tokens, and when we have a situation we need to respond to, any response or choice is limited to the available tokens. And if the temperature is too high, we end up with idiocy.] Everything in the brackets is a metaphor. Just a metaphor, not proof, not evidence of anything, just an illustration of LLMs that I've turned upside down, and said,'Hey, if you squint your eyes, that could be me!'
I think the most impressive thing is how the LLM managed to generate Ryan Gosling explaining this as though he didn't really fully understand what he was saying, he was just reading it off of cue cards that some LLM expert wrote. There are points where the explanation gets complex and you can feel the speech get a bit more hesitant and less natural and there are parts where the explanation is straight forward and you can feel Gosling all the sudden sound more confidence and fluid. Also, I like how he didn't really know what to do with his hand the whole time.
This is how all deep fakes look though, because the computer doesn't know what it's saying. It's spitting out words one after the other, but it has no understanding of the meaning of those words. That's why AI generated voices often sound so mechanical.
@@merlaak This is contrary to what I'm saying, however. It feels as though there is some kind of complexity and sentiment analysis being done by the AI because parts of the video, specifically the parts that would be easier for a layman to understand, come across as more natural as though Gosling had to concentrate less on making sure he was saying the right words and could focus more on the energy and enthusiasm of what he was saying. It could be my mind playing tricks on me and interpreting it that way, but it's how it feels to me.
Genuinely good video! Well written summary. Would looove to see a making-of video for this. Like how much was AI-generated vs explicitly scripted vs carefully curated, how much hand -holding and tweaking was required...
Honestly, maybe everything will come full circle now? If the internet cannot be trusted, then perhaps everyone will tire of their para-social relationships and touch some grass. I’m personally scared for the new generation of academics and graduates - imagine a world where your doctor used CHATGPT to write their thesis that got them their doctorate. I’ve noticed that kids are becoming academically lazy, it’s rare to see a child who actually wants to learn something.
@@jcozyytnah. Even that is going to be fake because people already Google stalk pretty much everything they can about someone before even talking to them. Plus with AI pins and other wearable AI it’s going to proliferate from there :(
@@JamesThomas-wf6qmthe question is whether you care ? Intelligence is using tools. If AI can pass the bar exam and med school wouldn’t you RATHER your doctor used it ? Good doctors are usually “good” because they’ve “seen it all” and can quickly access possible issues based on past experience THEY HAVE HAD. Imagine a doctor with the collective intelligence of ALL PATIENTS. You still need the doctor for certain things but offloading as much as possible is what they do already. Hence the specialities. Most doctors I’ve talked to can’t remember half the stuff they learned in medical school. They learn what they do every day and in their specialty. I’m not a pro AI fanboy but certain things will 100% be better with AI. Maybe it’ll knock doctors and nurses down a couple of pegs too :P Making healthcare more affordable and doing a much better job of triage :) I’ll add that it’s entirely possible with CURRENT TECH for someone to make a device to assess patient based on just visual cues and potentially suggest life saving treatments etc. it will only get better
Loved it, and there's something sneaky like a life-hack (potential risks here, but practically for this a positive) to getting your brain past certain barriers and open to the information right away.
The deepfake Ryan Gosling had his classic smolder going through the entire demonstration. In a way, this type of demonstration would have a dramatic impact on increasing engagement on a subject that is generally complex. It could do wonders for education, and this is a great use case from Synclabs and Elevenlabs. Bravo!
It is a basics, and I already knew that. Howewer, the klickbait Ken was too good to skip the video, and what is most important, the quality of the explanation was super high. Thanks, that was good.
This is SO well crafted. I think it's important for AI sceptics (like me) to appreciate that a video like this isn't produced in a couple of lazy clicks, but this was also thoughtfully planned and edited. Bravo.
This is maybe the best high-level intro video I've seen. The IBM videos are great, but if you have all of these different AI topics covered by deppfake celebrities, I think you guys really have something cool here. I think they'll be a hit. Going to watch the Chance one, now. Lol "tekst-summarization".
I watched... was going to click next... but then was continually intrigued, by both Gosling's voice and the well crafted content. I didn't understand everything but I think I got the gist of it. Well done! I wonder how RU-vid will transform as we begin to question reality. But would it matter from a philosophical standpoint. Would we care? Regardless of whether it was human or AI, generated, good content is good content. But maybe a more important question is the ethical implications. What if there are negative impacts to society that our promotion of it brings, like the disruption of job markets, lack of trust, and mass media influence and control.
The companies developing these AI tools don't care about any negative aspects like putting millions out of work, or eventually making hundreds of millions forget how the work is done in the first place. Once AGI comes around it is only a small step to conclude human aren't needed anymore. Humanity really IS abolishing itself and our political leaders and every other person is just watching it unfold.
Did Ryan Gosling give his permission to make this video? If so was he involved with any aspect of this video production? For example did he have to go into the studio and read a script or at the very least say specific words?
I didn’t expect much from this short video and its quirky format, but I have to say, this is a surprisingly good and in depth explanation of the inner workings of the transformer model and LLM’s!
Impressive, now I wonder what the real Ryan Gosling will have to say about it, considering Scarlett Johansson reaction to Open AI ChatGPT use of the voice Sky?
Well, this is just a short RU-vid video. They stole Johansson's voice in order to make tons of millions of dollars off of it without paying her for it.
People keep asking how the video was done but if you read the video description there is a link to the original video. So it's not a traditional face swap but a real video of RG with the audio replaced, lips resynched to the new audio and new graphics added. The elongated arm is a special bonus feature.😄
Or, you could take the opposite route and become an AI developer, understand the ins and outs of what you're afraid of, and get paid a lot of money. You might even build something helpful, like something to detect and track endangered ocean species.
@@user-ft2md2gd4s Woah broh, woah. That sounds like, man, something you had to think really hard about to come up with. I recommend checking out LessWrong and their concerns about extinction event concerns.
Ok hang on, how much is end-end AI and how much is editing? Did you generate the script with one tool, then input that to the others? And did you edit/stitch together the script and separate clips? I'm about to be flabergasted if this was all end to end from 1 prompt, even if it was more complex than the prompt you showed.
Hi livefromhollywood194. Sorry for any potential confusion. The intro interface was intended to show in a fun way that we're dealing with a deep-fake and is indeed not a real interface. The text was written by GPT, the voice was cloned with ElevenLabs, the text was turned into speech with ElevenLabs, and the lips were synced with Synclabs. We'll share a video soon with the full workflow.
It was also a bit an experiment how people would react to this kind of RU-vid format. Most of the tools used exist open source or have an API, so perhaps it might be a fun challenge in one of our upcoming videos to see how close we can come to the actual automatic deep-fake video creator by leveraging some of these open source tools, API's, and autonomous agents. Challenge accepted!
As a layman, I have a pretty good grasp of how LLMs work, but the last 10-15% that this video filled in for me gives me a much better understanding. While I knew the text generation was performed essentially word by word, I didn't quite make the connection that tokenization essentially means the embeddings are generated at that level as well, and the example explanation of what an embedding might represent really drove it home for me. I should have bothered to learn more about vector embeddings sooner--they're way more interesting than I gave them credit for. Also, the analogy of the sequence of events in one's life predicted as an LLM might predict the next word in a sentence in comparison between pre- and post-transformer architectures was a stark but effective one. I definitely have a richer appreciation for the All You Need Is Attention paper now. Thanks deep-fake Ryan Gosling!
That woud double-interesting! Not only was the deep-fake real, real good (maybe apart from the mouth, sometimes), but the lecture was real informative and straigth forward. Awesome! Thanks!
This is actually a very good explanation. Have to ask why Ryan sounded so sad thinking about high temperature factor resulting in gibberish, like he's truly felt that pain