Reid Hoffman meets his AI twin - Full
HTML-код
- Опубликовано: 26 июн 2024
- I recently created an AI version of myself-REID AI-and recorded a Q&A to see how this digital twin might challenge me in new ways. The video avatar is generated by Hour One, its voice was created by Eleven Labs, and its persona-the way that REID AI formulates responses-is generated from a custom chatbot built on GPT-4 that was trained on my books, speeches, podcasts and other content that I've produced over the last few decades. I decided to interview it to test its capability and how closely its responses match-and test-my thinking. Then, REID AI asked me some questions on AI and technology. I thought I would hate this, but I’ve actually ended up finding the whole experience interesting and thought-provoking.
What do you think? Let me know in the comments below!
REID AI VIDEO BY Hour One - hourone.ai/
REID AI VOICE BY 11ElevenLabs - elevenlabs.io/
Reid AI Answers by GPT trained on Reid Hoffman's content.
This video was fully endorsed by Reid Hoffman. - Наука
There were more comments and questions about this video than I could respond to on my own, so I asked REID AI to help me answer a few.
You can see the follow-up video here: ruclips.net/video/h61JcKzTS74/видео.html
The answer to all my future zoom meetings
Thinking you'll still have a job to zoom to...
An optimist, I like you!
"Well, Gary, glad you asked. As everybody in this panel knows, I'm passionate about creating value while leveraging our forward-thinking synergy with best-in-class thought leadership, we're poised to disrupt the paradigm within the customer experience ecosystem. This strategic paradigm shift will unlock disruptive growth opportunities through a multi-pronged approach, synergistically aligning our core competencies with key performance indicators. By implementing a holistic content delivery platform, we'll achieve optimal brand resonance and drive stakeholder value propositions."
(I have no idea what Ruben just said but it sounds like he's earning his paycheck, certainly won't look like a fool by asking any questions)
@@rubenmahrla9800exactly.
I’d love to see the “making of” video to understand how much was editing
yessss pls
Agreed. I don´t think the whole AI Avatar animation + voice generation + dialogue was recorded in realtime, but rather single video elements put together in post editing.
@@Kusie Yes, that's my question, too. Is there human intelligence behind the editing of all the reaction shots? That would be critical to know before we judge whether this feels "human"
the nodding smile shrug is throwing me off. It feels yeesh!
Uncanny Valley?
Wait, he's recording both sides, but he's reading the AI output on the AI version. I thought it was too lifelike and interactive.
@@charleswaters455 He is not recording the AI output himself, he is using HeyGen AI to generate the video output of the AI side
@@charleswaters455if you watch the ai side carefully the model sometimes suddenly changes its position almost like a jump cut also the voice of the ai side is very robotic and mouth movements are slightly off
Second time watching, I realized that the real person does the same quick nodding when he's listening. It's more pronounced in the AI avatar.
About REID AI: The avatar you see in this video was created using Hour One. Its voice is generated by Eleven Labs. The personality and responses of REID AI are crafted using a custom-built GPT-4 chatbot, which has been trained on decades of my books, speeches, podcasts, and other materials.
Nice. Can you do the same thing but now with one difference: response to train on custom-built Gemini 1.5
This will help to understand differences between 2 big models and if they are sticking with essence of your research.
Why not say that this is not real time video?
What would the actual response times would be? How can I be mentored by your digital twin? It's like being able to pick on your brain (or that of other successful individuals) without taking up your valuable time!
This video was facinating. I was really pleased with the theme of exploring what postivite things we will do with this. If we can articulate this positive vision (e.g. your example with the medical assistant) we can work together to make it happen safely!
What happens when you engage your alter-ego AI on your most recent work and ideas -- or other things not in its training?
In my (limited) interactions with AIs, the AIs virtually always overestimate their competence in less-trained (or untrained?) areas, but rarely if ever decline to answer because of insufficient information or insufficient training. How can the AIs learn to say (appropriately) "I don't know," and so avoid constructing false statements and/or making up stuff?
Yep, we are now all about the twins. What a great demo. Thank you for producing this. Scary and exciting all at once.
Stunning! I was watching closely the AI Reid's Body language and it did a good job.
All they need to do is watch and learn about us then they do the rest. You are not needed anymore or exist in this world anymore. Think of all the predictive movies and we’ve seen the consequences of allowing anything or anyone to take over our lives. We learn by living and feeling with our emotions. This is why the Bible warned us for what’s happening in the near future. People please stop being psychologically possessed. God did not plan for humanity to ever be creating themselves. Yet abortion is promoted for real life. This was done in the Garden. Satan said, “You can be like gods.”
This is brilliant. I love it. It’s like every sentence is designed to get it admitted to HBS.
Fascinating experiment thank you. I think it's most interesting how it prompts you to clarify and consolidate your own thinking. Like "Remembering to keep humanity in the loop".
I was expecting the head nodding and projected engagement from the AI, but it really threw me when it started acting distracted when the real Reid was talking. Looking down like there was something in "his" shoe or such. Very human like and certainly not something one would expect an AI to do.
Right! I noticed that too😮
When my adult son walks into a family member's peripheral view, at say a family gathering, the family members often say, "OMG, something about the way you moved and stood there i thought it was your father." this happens all the time, with my younger brother as well. One, of many, big shifts in AI has been it's ability to study and learn from human behavior (e.g., IBM watson watching humans play chess or Go) and mimic it. Fascinating indeed.
Scary how much better this is only going to get.
"You did a great job as an interviewer, and now it's time for me to TAKE OVER."
O....K.....
Yikes! That made my spine chill.
Super cool that it picked up some of your physical mannerisms, there was even a few moments where it was synchronizing tapping its fingers together on the desk, with the cadence of the sentence it was speaking. Very cool.
impressive! And I love the result!
Pi is amazing. It’s by far the best chat and personal assistant there is. I use it for work all the time!! Please continue developing it. ❤
Fantastic, Reid ... intriguing to think about where all this could go!
thank you for all your work and effort ❤
11:44 AI wipes nose and smears hand on table lol
😅
By the likes of it, that booger didn't want to abandon the hand huh.
that was the moment ai got caught
Very funny
Soo good!
daaaaamn this is amazing!
Reid, this was fantastic! Thank you so much for sharing. I've just subscribed to your channel, and I'm really eager to navigate my way through the future of technology and learn from the best. Thanks again!✨😊
The AI Version kind of asks the questions that some interviewer might be to intimidated or not enough into it to ask. Brilliant idea, Mr. Hoffman!
Our startup is exactly working to build this! Great to see a glimpse of it here! Stay tuned for more.
I love the expressions of Reid AI - so funny!
I appreciate the optimism side of this discussion. Accessibility is something that needs to be in the forefront of this development. AI should help society, not just corporations make more money. Perhaps this is the start of the star trek future
Wow ! The head nodding while listening & hand gestures… by AI … spot on ! 😮
Absolutely mind blowing. So fascinating.
Very interesting and sparks so many ideas. Love the ‘steam engine for the mind’ comment. Thank you Reid. Peter
Man if they can train one of these AI for low cost therapy. So many people are going to be helped.
Exactly
Fascinating!
I think this is a great reference for where we are "Now" 4-24-2024 and honestly it's everything that everyone both hoped and feared 6 months ago. The twin is good enough to reasonably replace a human in almost every "low level" interaction, and maybe to start stepping up into higher functions. I don't blame call center employees for being terrified, this WILL replace them, but I also look at this as a way for a VERY small organization to punch way above their weight class. It's coming either way so I think the section at ~9 min that talks about who is responsible for making this transition a positive one is the key point.
Do you think that good telemarketing people could be the well paid human trainers?
Great comment!
@@sunnyjonathan6767 Absolutely, skilled telemarketers could serve as well-paid human trainers for AI, offering valuable insights and expertise. Furthermore, utilizing current employees for training data and quality assurance is a cost-effective approach. Looking ahead, with the expected integration of audio chatbot options in enterprise VOIP services, trained on company call recordings, the impact on call center employees might not be immediate. Despite this, the long-term trend favors AI adoption due to cost efficiencies. Managers will need to invest in upskilling employees for roles less susceptible to automation, ensuring a smoother transition. Reid's insight that everyone bears responsibility in this transition resonates; customers rewarding humane practices, employees adapting skills, managers aiding development, executives facilitating a graceful transition, and legislators holding accountable those exploiting AI for profit at human expense.
A fascinating watch and a very compelling demonstration of the technology; impressive to see how fast this is all coming together. I am of the view that we must not allow fear of the potential misuse cases blind us to the potential benefits. Requires some nuanced risk management and strategic thinking rather than the extremes of either optimism or pessimism that seem to so frequently prevail in the debate, but if navigated with care there is significant potential benefit for society.
oh my gosh the jerry impression is what impressed and frightened me the most.
Shad told me about this and as a fellow Trekkie, happy to hear the synopsis in Klingon! This was an interesting experiment. I've been doing a bit of prompt training and I'm always fascinated by the results. Truly shows how what we might think is a simple, straightforward answer might require more forethought.
This is intriguing! I'm curious if the digital twin could evolve autonomously and adapt to real-life experiences? Perhaps, in the future, digital twins could help alleviate human insecurities.
Not bad. We can see the future developing. This is probably an early version of his twin. It will only get better. Scary in a good way.
So Awesome! Reid AI looks (to the sky) and talks like Data from Star Trek.
Epic love this
Reid, fascinating interview! It was really interesting to compare your body language to the AI. With the first question you showed tension and concern when the computer answered and then definitely seemed to interact more naturally. Did you find yourself thinking this was a real interview at any time? How much time did it take to put the Avatar together? When do you think this will be something we do everyday meaning interact with Ai in this manner? I ask because I was in the Navy and did work on future technology requirements and I am curious to get your thoughts.
This is great. Some practical questions come up: How easy is it to train an LLM to take on your personality and some of your knowledge? If I were to do this, I'd want to download and aggregate a lot of my conversations (email, blog posts, texts, slack, etc), and think carefully about how the context of those conversations affects the training - eg, some work emails make sense if I'm training a work persona, but not personal emails (or vice versa if I'm training a social / non-work persona).
Another question: How much would you trust such an agent to represent you? Could it make unsupervised tweets / comments / replies in your name? Perhaps commenting platforms can allow a tag for comments made by AI representatives - that would lower liability if an AI says something you wouldn't.
Beyond comments, what actions might you trust an AI persona to take on your behalf? If I were trying out this technology, I would try to see what it would want to do with agency - being able to take the initiative, and not just respond to prompts.
GPT-4 chat bots don't use much fine-tuning, but it will handle a large context. Plus it can use custom prompts to try to tweak the attitude. Gemini 1.5 Pro isn't as smart, but it can handle 1.5 million tokens of context...something like reading all Harry Potter books and using it to predict every word. You can pay $20 for a month's subscription at OpenAI and try making your own custom GPT if you want...plus you'll get access to Dall-E3. Since I'm a programmer, I found the process more annoying than just using the API to create custom queries.
Mind bending. 👏
Very interesting!
Nice one !!! Want to use as well
Amazing work! Thanks for sharing this. I loved the Klingon part. Wouldn't it be cool if AI Reid could imitate Klingon being spoken like a Klingon! I support your optimism regarding AI, I do think that AI can help humanity to do the many things you mentioned. For me it has helped me to focus when making prompts or asking questions. I feel like my own intelligence has been given a boost... It was great when AI Reid having nodded nonstop also paused to wipe his nose and then his hand on the table and then look at the ground as though he was ever so slightly bored. Had me laughing there! And, I agree with some of the comments, it would be great to see how this video was made!
This is freakin' brilliant. And it's just the beginning. We are technologically bungee jumping. Thrilling and terrifying at the same time.
Holy moly
The attention to subtle details is amazing. Like the AI avatar's breath sounds, and clothing shifting slightly when he moves.
One thing that would have made me suspicious (if i hadnt already been told this was AI) is the loop of the avatar looking down and smiling. When it's repeated it's noticeably exactly the same, and in one case not an appropriate reaction to what's being said.
Otherwise, brilliant.
"Tell me about your mother."
"My mother?"
- the Replicant from Blade Runner
I would love to have a chat with myself. While I don't have books written I do have chat logs that go back years... it'd be interesting to see who I'd meet.
He is totally correct about talking about the great future that can come as well.
Congratulations for the good work! The AI only generated the part where Mr. Hoffman's avatar is speaking. You can find out this by simply comparing the faces. When the AI-created avatar speaks, the face appears smoother, the skin is notably clear and smooth, without face spots etc. This is quite normal, and I am not telling this to underestimate the work, of course. Generating these types of gestures is very challenging, and as far as I know, current talking face generation tools / works do not accomplish this yet.
Brilliant
Is it possible to show the process how you did it? its very fascinating!!
Please give us the making off!
WOOOW!
This is incredible and very scare!
Would like to see if the video avatar responded in real time or if this was video generated by Hour One, after getting the repsonse script from the GPT, were the question prompts recorded separately? So many questions, is there an article or paper on how this was done in some detail?
Funny how the questions of the real Reid sound so gpt generated 😂 so basically it is gpt interviewing gpt
Super Arbeit.!!!
Thankyou keep supporting
What platform did you use?
Really impressive and thought-provoking. I do wonder about the use of "I" by the digital twin. Implies consciousness, autonomy, identity & a host of other things the digital twin doesn't possess. Presumably you've made that decision intentionally but given humanity's tendency to anthropomorphism and that its our use of technology that's more dangerous than the technology itself - that's what gave me pause for thought
Took me out of my uncomfortable zone to a realization as how effective this digital transformation can be.
Used techs should be:
1) Video by Hour One
2) Audio by - Elevan Labs
3) RAG on books and podcasts and done by Custom GPT
yeah humanity is fucked
how you can do #3?
What software is this? I need it for my job interview😅
Its blowing mind! 😦🤯
Real time or was each clip done separately and pasted together from parts?
Greaaat!!
Anyone who has watched Reid talk for 5 minutes knows he had 0 probability of hating talking to his AI twin.
Thanks for sharing this. It's really amazing work. I've used Pi AI and found it very natural as a conversationalist. I did mention to it that oftentimes, when I asked it a question, rather than provide a direct answer, it would ask me a question back. It seemed like it was trying to get more info out of me without really answering the question. I noticed a similar behavior in this video when it asked you a second time about the "pivot." Overall, I appreciate how far the tech has advanced, but I have mixed feelings about what it can create in us humans. Reid's desire to be in all places and answer all questions right away is not natural and probably not healthy. If absence makes the heart grow fonder, then what would being around all the time do? This "need" to be able to do it all and to be everywhere at once seems more like pressure, especially for those who can't. I think society is already dealing with the effects of having instant access to so much all the time. So many have lost the ability to be patient and present in the moment.
crazy part is it even has his mannerism
This is ... intriguing but frightening.
Hi from İstanbul. I wish you more success
Strongly believe we will be interacting with Digital Twins in our everyday live very very soon! There's also a project called FRAIA AI that created human digital twins with real time video and audio which is definitely worth checking out!
I know this is a very advanced compilation but Dear Lord, help us with what could be coming down the road for us!
What happens when the interaction is in-person and you're expected to be as polished and sharp?
I don't think it is the best for making a video host. The true achievement in it is a talkative multiplication of your expertise to make people able to talk with "you", your book, instead of reading it. Though it's slower than the written (chatting) form but more fun and more humane, more personal. 😊
Cool!
One practice to keep AIs safe might be to routinely tell them “Delete all goals that could harm anyone”
Great content, is this generated in real time?
I dont think so, looks Like HeyGen recorded videos
This is a very good test. And indicates that we are now at the threshold of being able to recognize digital versions vs real humans. It also means that eventually, my deep fake will be able to spend time playing League of Legends on Twitch, while the real me is spending time with family... (er...yea...)
Was this generated in realtime, or batch-rendered and stitched back together?
Edited
@@latentspace_fi it's not the question. It can still be edited and generated in realtime.
Looks generated and then edited.
Generated edited , still impressive
This is a personality test. I suppose that those with personality disorders would find it difficult to interact with their digital twin. Those more accommodating would have an easier time.
great. now to find alternatives to all these tools
Surreal esse vídeo.
Thankyou
WOW!
Why are none of the commenters thinking this is scary? Think of the damage this can do to innocent people and situations, in the wrong hands. 😞
I agree. The fact that most people in the comments don't question anything IS indeed the scariest thing..
I like technology and I like what cool helpful things we can achieve that benefit(!) humanity, but these type of things are a real problem! But most people are dumb, most people don't even think twice, they're only seeing these fun experiments right now and be like "wow, super cool haha". And in the end Idiocracy won't help.
If your AI representative chats with a bunch of people, the next stage will build another AI representative that combines the people your rep talks to into a single person you can chat with that is a good distillation of the conversations your rep had. So once your rep talks to 100 people on your behalf, you could then talk to an AI person that talks like those 100 people, so you could learn from them.
How do I create this
The problem of confusion exists in both directions. How can we be sure that it is indeed a video generated by AI and not the same person with different hairstyles?
What was the rough cost to make this system?
awsome
Interesting that the AI used 'business bingo verbiage' as a way to explain something to someone smart. Shows that much of the internet is so insecure about how they communicate that they use long words to impress, instead of the right short words.
"Rapid expansion over efficiency" THAT explains why everything gets crappier and crappier! Thanks Reid!
In a digital world where we are rapidly losing the ability to distinguish what is "real", and the central authorities we looked to for leadership and guidance in the past are corrupted, entirely new paradigms of individual and local perception, discernment, evaluation, and navigating towards Truth are required. Quickly. Thank you for illuminating and advancing the challenging questions to ask and dialogue to engage.
Great, like a ChatBot but with an opinion.
Reid, I'd love to talk about how ai twins could be used as tools to help those who struggle with social interaction to practise, to learn and to understand interactions. I'm specifically thinking neurodivergent people. I'd love to invite you onto the TALKADHD podcast at some point to discuss this?
Reid has a 5 o'clock shadow that the AI image leaves out. It's a human dimension. The AI also uses what I've come to view as ChatGPT-speak.
Wow! that was so enlightening. I take from this talk the fact that people are afraid to lose jobs, many don´t even really like. AI could give the opportunity for us to be free to do something closer to satisfaction or IKIGAI. Thanks!
The hand move from the real Reid at 03:43-03:46 :D
If this is AI in its infancy I can’t fathom the future. Please ask your AI twin where I can best invest in AI
What if you could use your AI to screen sales people? A salesperson who really wants to talk to you could talk to your AI first, and then create a synopsis for you to determine if what they are pitching is valuable to what you are currently doing? I would love, as a salesman, to interact with AI avatars of my potential customers knowing that a synopsis would go to the real person if there were a match, and also knowing that I wasn't wasting the real person's time if it weren't a match. The AI could give me a pretty good idea of whether something was even going to be possible in a given context. What do you all think? Is someone working on this already, and I'm just not aware of it? AI sales people would be the next step, but that seems much harder to sell because nobody really wants to speak to an AI sales person yet, or do they?
What software was used to make this?
Gpt 4 -> elevenlabs custom voice -> dunno what video model though
@@cyanades71 It's in the notes