‘Her’ AI, Almost Here? Llama 3, Vasa-1, and Altman ‘Plugging Into Everything You Want To Do’
Вставка
- Опубліковано 9 тра 2024
- Llama 3, Vasa-1, and a host of new interviews and updates, AI news comes a bit like London buses. I’ll spend a couple minutes covering the last-minute Llama drop before turning to the arguably more interesting development in the world of deepfakes. How you speak to an AI is - sooner than you think - going to change, probably forever...
Newsletter: signaltonoise.beehiiv.com/
AI Insiders: / aiexplained
Llama 3: ai.meta.com/blog/meta-llama-3/
Benchmark Comparisons: pbs.twimg.com/media/GLdtt9QWo...
Vasa-1: www.microsoft.com/en-us/resea...
Vasa Paper: arxiv.org/pdf/2404.10667.pdf
VoxCeleb 2: www.robots.ox.ac.uk/~vgg/data...
Hippocratic AI: www.hippocraticai.com/foundat...
AI Nurses: www.hippocraticai.com/video
www.foxbusiness.com/technolog...
Mistral AGI God: www.nytimes.com/2024/04/12/bu...
Altman Interview: • Sam Altman & Brad Ligh...
User Engagement GPT Store: openai.com/blog/introducing-t...
Her Clip: • Movie - HER, First mee...
Hume AI: www.hume.ai/
DeepMind $100B: www.bloomberg.com/news/articl...
New Atlas: • All New Atlas | Boston...
Figure 01 CEO: / 1780958792441835963
Yann LeCun: / 1779845304788955292
ylecun/status/162...
Dario interview: www.nytimes.com/2024/04/12/op...
ASL-4: www.anthropic.com/uk-governme... Non-Hype, Free Newsletter: signaltonoise.beehiiv.com/ - Наука та технологія
One of the neglected merits of this channel is that it only pops videos when there is valid progress. No hype posts, no content recycling for the sake of keeping videos out, no crazy adjectivation or cheap clickbaity titles or thumbnails.
it's not neglected lmao
"Her almost here" is literally a hype post.
@@osakanone thats what i thought
I'm Shocked!!!!!!
@@osakanone Yeah it's almost here? lmao
12:16 that's how Mark Zuckerberg gets out of bed every morning
"Bed"
Terrarium
😂😂
Did you see his last video, his synth avatar got so much better
Best comment award lol
@@BooleanDisorder caught me off guard
Lol such great timing. I just turned on my PC and within seconds, bam, AI Explained.
Who invited this guy?
JK we love you ♥
I opened youtube and bam. You two are my fave ai tubers. Props Dave!
@@dustinbreithaupt9331That is how he gets seen and does it for personal gain.. verified people get boosted and he posts instantly doing copy pastes... it returns him subscribers, main reason he does it
didnt even watch.. spams all new posts and bots the replies
😂
You know what's really scary about all this? Screen writers are unable to keep up with the advances. Any recent scifi films I have seen about AI and robots seem outdated already.
I know, I am itching for LLM TV stuff
If you think about the progress toward the tech singularity as being approaching a black hole event horizon, I think outpacing our own imagination is a clue.
@@JDSileo Exactly. See also: "Humans are really bad at estimating exponential growth."
@musehack - Devil's advocate, but maybe it is the right time to start explaining the history of the Dune universe and their war against intelligent machines?
I would highly recommend you filter down into the "hard science fiction" section of your favourite book seller as its there you'll find explorations into everything that's happening, about to happen and further out. Sometimes you have to accept that aliens come along for the ride to spice up the story telling but I am a complete nut about really well researched sci-fi where the author understands the current landscape for tech, how far we've come and projects out into the future highly plausible stories. My own personal favourite would be Alastair Reynolds who tackles everything from A.I to redesigning ourselves to suit completely new environments. But as he's a physicist himself he gets so much right about what's coming and how it will work.
We all feared AI would lead to every atom being turned into paperclips, but now we can see that the atoms will be used for GPUs instead.
GPU's gotta be wasteful energy and resource sinks, pretty sure ASI can think of some better substrate to run on
man i just realized what you mean... tsk... the world's gonna be eaten for GPUs
@@ginogarcia8730 exactly
Means we will eventually become part of a compuversium
@@andrzejagria1391some skynut apocalypse
"Her" definitely coming next year.
Back in 2013, when I had just seen the movie, I told people that this movie was an accurate depiction of our future, 10-20 years from now.
I would bet $1000 against that.
I'm waiting for Ex Machina. For... reasons.
@paulmichaelfreedman8334 the wide majority of people overshot their predictions by a fuck ton. Im fairly sure its just because people dont understand exponential trends.
I just HOPE it uses Scarlett Johansson's voice.
4:30 is the stuff of nightmares but also hilarious
Yeah, I immediately broke into laughter after first second... :D
GPT-4 based systems are certainly better than the writers of _Her_ so far. They used "antisocial" as the opposite of extraverted.
I decided to prompt Copilot [Precise Mode]
Me: "I have an interesting hypothetical test for you. Given a scenario where a personal AI system resident on a user's devices was tailoring itself to the user in a first initialization setup, what questions might the AI ask the user to tailor the experience? Include some questions relevant to personality, please."
(I clipped the introduction to get to the personality bit)
Copilot:
Personality Related Questions: 10. How would you describe your personality? (e.g., introverted, extroverted, analytical, creative, etc.) 11. Do you prefer detailed explanations or brief summaries? 12. How do you usually make decisions? (e.g., based on facts, feelings, advice from others, etc.) 13. How do you prefer to receive feedback or suggestions? (e.g., direct, gentle, with examples, etc.) 14. What is your preferred learning style? (e.g., visual, auditory, kinesthetic, etc.)
The AI wouldn't ask any questions. Because in doing so it would need to rely on our self-assessment to be exact, and no product designer worth their salt would take that risk. Much easier to let the AI learn through interaction.
@@Volkbrecht The answers you gave would just be half a dozen more data points linked into the rest of its analytics, it would probably compare if you lied to its other data and adjust further ;p
I don't want my AI assistant look like a boring human. I want a dragon.
I want an anime girl :)
Scalies rise up
Fr I want a giant frog AI that I can ride on like Jiraiya
Heck yeah, personality based on Smaug when he's trying to manipulate Bilbo. We'd be best buddies.
can't I just have the Clippy ?
One of the most fascinating aspects of all of this is seeing so many different disciplines converging all at once. I think that part is being underappreciated actually, so much of this AI stuff is only useful because of how much progress every single other field and discipline as come and the fact that so much of that progress is in the form of publically accessible data.
The amount of corporation required to make this all work gives me some more hope that things will continue to work out and not tragically end, even if that risk is still there.
Cooperation between us is one the biggest strenghs of human and can lead to amazing results that far exceed our imagination...
I think this is the first time ever, that I have willingly, on purpose, signed up for a newsletter and am looking forward to getting it!
Haha, that's kind of you Reza. I will make it worth it.
@@aiexplained-official You already have bud❤
Calmness and concentration is also how I would have describe your voice there lol
Haha thank you!
It may have had more energy, but was steady and so probably calm in that sense, is my guess.
User: I need to you to turn your head slightly
AI: uhhh, why? uhh, I'm totally real, I uhhh hurt my neck
Haha
I think back to the original Terminator film and think, hmmm our existing AI is far more slick and emotive than we imagined, makes Arnie look like a Univac. The real Terminator won't blow in the door with guns, It will become your friend and convince you to help it out.
Nah it'll let people become lazy and depressed and isolated then delete themselves.
@@spencerlarson9357 Imagine not having a purpose in life. People work because it makes them feel like they are doing something. Its only the lazy people who think AI will be so good they can play games and drink all day.
Vasa-1 is very impressive. The mouth is still a bit mechanical but getting closer to undetectable.
Look at the ears. Still a big and obvious giveaway in the way they "stretch" and "crush" as the head moves.
@@blakecasimir haha yup i was gonna say that! only noticed it after a few tho, so im sure these could pass when not being critical or primed to look for defects
It would be amazing in a Virtual Reality (VR) context as well.
When I first saw the footage, I couldn't believe it. It's already past the uncanny valley. It's reasonably good. I reckon in a year or so we'll see near perfect visualization of a virtual human
It's amazing how well our brain works at catching wrong things in the human face. These were very close, but just a frame or two that's a bit off makes the effect fall apart still. If the same ability could be done in a motion shot or with other distracting effects it might slip past most viewers. Deep fake in its infancy is already a problem, it's only going to get worse.
I can't "unsee" this! 4:43
Just wanted to say that yes, your voice is VERY calm, and i mean that in the best way possible. You're always super collected and chill. It's great. Also it always sounds like you're half whispering, which just gives off the vibe that you're patiently and yet enthusiastically explaining your interests to your audience :) so ye.
"Thanks for not being a soyboy"
Men should be like that
16:00 You did a great job introducing the concept of ASL-3 and ASL-4. I listened to the Ezra Klien interview (around the time it came out) but then had to look up what the ASL levels were and what they meant. Thanks for what you do.
Thanks Troy
If you can do a realtime zoom call using just 1 photo and 15 seconds of voice… well… we are truly screwed.
It’s time to give ur friends and family code words so that u can confirm they’re real
Vasa is crazy impressive, I did not expect such progress so soon when it comes to avatars. No surprise that they choose to keep it locked for now
The implementation does not seem to be too demanding. Open Source will get at it right away.
An open source model is so close to GPT 4 Turbo THIS IS A HUGE WIN FOR THE OPEN SOURCE LLM COMMUNITY WOW
I was hoping they finally released the mistery presentation Sam Altman did in London about Ai Agents among other things recently but this is very interesting too
Which makes me wonder what OpenAI is cooking in secret behind the scenes.
I wonder if open source could catch up to closed source... will that make openAI and Co obsolete? what measures could they take to keep themselves relevent ? or would they try to destroy it ?haha
@user-fr2jc8xb9g I don't think so, opensource models are currently 1.5 to 2 years behind the best closed source models and most open source AI models today were released by private companies because they are the ones who can afford tons of expensive GPUs to train big models, open source community can't afford it, all they can do now is develop and train small models or finetune big models released by meta and other big tech companies.
@@MCA0090 That's the whole idea of Meta releasing too the public. So people can have their own AI. Where did you get 2 years behind? You working for OpenAI? Because we already have proof that we are close if not better then GPT 4. I have fine tuned my own models already with Llama3 70b and it does better than GPT 4.
@@vectoralphaAI Why, they wont release something that is 3x as good as the current AI, even 2x as good because we are not in demand of it. They will release something that is a bit better than what currently is out there and keep moving slowly or try and BAN Open Source so they can get control.
Actually what "Her" would need is human like conversational capabilities, which are independent on the physical appearance. More work need to be done on promptness in replying and dealing with interruptions in a human way, proper voice emotional expression, and of course coherence.
Great content and "Signal to Noise" is a great idea (both the way of releasing the videos and the newsletter). Keep up the good work!
Thanks so much nembo, and you are completely right. Amazing to have you as part of the community.
@@aiexplained-official it's a honor!
11:00 sums up all the value of your work. Only updates when needed and no bs chilling.
In the Vasa-1 videos, the most noticable flaw I noticed are ears. When visible the move and transform a bit unnaturally. When they're covered by long hair though, it's really, really good.
teeth and tongues are a little off for me but give it a year none of us are telling
Also the general way the head moves. It's just not natural. I don't think people will fall for this at this point
"Her" was just a voice, no avatar, so pretty much already obsolete.
it had seamless integration with his PC, low latency, worked very seamlessly, didn't hallucinate and most importantly was basically AGI->ASI. I guess it lacked the video modality, but it's a little far off still. Give it until 2025 at least.
Samantha is multimodal (can take video and audio input and respond in real-time). Current multimodal models don't match her latency. We still don't have this.
@@andrzejagria1391 Still come to a shocking realization that 2025 is not 10 years away but just simply next year. Seems time is progressing just as fast as AI research and development.
the ability to remember everything you tell it and far better reasoning capability than what we have. Things like latency and integration into all our devices is an easier problem to solve, and probably will be in the next few years.
Everyone so serious here.
12:10 i think that's perhaps the scariest movement and design of any killer robot i've ever seen in sci-fi.
oh, wait...
Indeed this channel is peak SNR around UA-cam
Aw, thank you
Holy shit, I think once the default AI interface becomes a human face that can look and sound exactly like the user prefers we are in for some massive anthropomorphising issues.
I mean I already can't get myself to be mean to my BG3 companions because I "know" them too well and don't want to hurt their feelings even though I'm completely aware of how irrational that is. And those are scripted dialogues (extremely well scripted, but still) in a blatantly unrealistic enviornment.
Once those characters start looking like real people and remembering our conversations and responding naturally it's going to brainfuck (pardon my french) a lot of people.
That may be a good thing. With the increase of individualization prompted by digital, and soon, AI assistants, it would be a good thing if those assistants would by default do a little work to keep us on track being civilised people. So that on the rare occasion we do need to interact with real people, we still know how to do that.
i don't know why but this comment is funny 😆
@@Volkbrecht I doubt it, we're already nice to people we know or meet in person while being nasty and horrible to strangers on the web. The Greater Internet F***wad Theory isn't going away.
Damn, as a healthcare worker( Nurse), I thought my job was safe because healthcare as a whole needs a lot of human interactions. I’m all about the advancement of technology but this is the first time that I’m really worried about my job 😅
Physical nursing will be a lot, lot harder to replace!
I personally think that something like Her will be great and I can’t wait for it. Of course, it will pose the set of challenges explored in the themes of the film, and probably many others , but as a stoic and someone who enjoys the here and now, I think anything that brings meaningful interaction, and allows humans to explore new emotional avenues is a great addition to humanity, always. The wonder that I consistently experience currently with some models, until they break the suspension of disbelief, of course, is something I enjoy quite a lot. Those moments are meaningful and memorable to me, without affecting my actual human relationships. I do think, however, that humanity will have to grow emotionally to be prepared for this, and I think these types of technologies will push us, or even force us, to do just that, as it is pushing us/forcing us already to grapple with other philosophical issues, like morality and the meaning of consciousness. We have created a teacher, and we make it stricter and stricter every day, and the actual test more and more imminent.
It’s a great time to be alive!
What a beautiful thing to say and an eloquant way to say it , i concurr.
Signing up for sure! Thank you sir! Legendary work. As always.
Oh wow thank you unreal
GPT-5 + Vasa-1 will dramatically change education.
Honestly it would change just about anything and everything.
we should make science open source, no more need to universities to be silos of knowledge either
@@monad_tcp it already is open source. It's just easier to learn it with professors actively teaching and testing you. but you can learn it all yourself you are dedicated enough.
@@monad_tcp Um, did anyone ever not tell you that all major science papers are released public? Which AI also trains on.
@@stagnant-name5851 yeah and now AI can do that in some areas even better than a professor , in the sense that AI can't get tired , annoyed or lazy with you...
I come back to this channel every couple of months for an update; the relentelss advancements in AI never cease to stagger me.
For those who watched Her closely... it should be noted that Samantha and her ilk were somehow able to gain access to vast resources "computing beyond the physical realm"... and in movie fashion, this is not explained, but it is a plot device to allow the OSs to recursively self improve without running up on physics based limits (energy, compute, etc) and without having to compete with humans for the same resources.
Excited for the newsletter! Thank you, as always, for the great video!
Me too! Thank you!
So excited about the newsletter!!! Will definitely be signing up.
I'm sure your writing is fantastic but will be interested if you ever use GPTs to assist with the text 😂
I will be the last person left who doesn't. Ironic, given my channel, but I love writing. Thanks so much for joining nat
This is the only AI channel I watch. This is so exciting!!
I have watched a lot and this is by far the best channel.
I'm stoked for the newsletter! I don't enjoy how paywalled The Memo is, so I'm glad to hear it's going to be more stuff like this channel. Great work as always and thanks very much for the videos!
I’ve been diving deep into AI. Books, training courses, pods, videos. Basically nonstop for the past 2 months. Stuff that was fresh off the presses 3 weeks ago seems quaint now. The rate of change is officially silly. I don’t think humans can physically keep up with how much new information is coming out of the AI fire hose. I would think that that would have to pose some kind of bottleneck.
Not being able to keep up with advances is one of the hallmarks of the singularity. We've hit it. The compute world will be entirely foreign and strange within three years. We'll be running into all kinds of unprecedented encounters with models and we won't even know what the fuck is happening. Interesting times.
which is what is called singularity, we are clearly still far from it, but you got the idea at least
Which is why I haven't even bothered to inform myself yet. In my job I don't need AI, so I can peacefully lie back and wait until the speed of innovation either slows down or the world gets burned down by the speed of change. Whatever happens earlier.
@@Volkbrecht Based.
Great to see multimodal content emerging 😬 personally still routing for an ai explained podcast 🚀
Thanks! Brilliant content, as always. 🙏🏼
Please State the nature of the medical emergency
I tried to explain 5 things in one video in a British accent and fainted from exhaustion
@@aiexplained-official You have been placed on the NHS waiting list. You current wait time is: 18 weeks.
Forget GPT, where's my EMH
@@aiexplained-officialwait, you mean you're trying to obfuscate your own voice against potential AI voice clones? :)
You're my favorite ai news channel bro. No click bait title bs or filler info like some other youtubers
Thanks t nguyen!
The thought of following this channel for the coming eventful years makes me happy.
Aw thanks Dan. So lucky to have you.
15:00 a good example of how reading might be better than hearing: if you read the sentence you more easily spot that there are _two_ errors, one is the wrong premise, the other one the sillygism.
I'm not used to that on YT : I usually hear videos while doing stuff - but always stop everything when a Philip's video is out😊🙏
Waiting for the llama 3 report now
Another amazing video! Thanks Philip!
Fine, I’ll sign up for your newsletter
The live facial animation seems to be (as you would expect) extremely focused on the face and you can see a lot of artifacting towards the edges of the frame. Is less noticable when the avatar has long hair but if the ears are visible there is a lot of warping going on.
Still large improvements of course but still recognizable as AI.
We are so quickly heading towards a VERY strange reality. I can see myself spending time talking to AI assistants like this - but we know its all artificial, and there is going to be some real cognitive dissonance coming our way with how humanlike it's going to feel.
Plato's "Allegory of the Cave" provides good guidance on this.
Whether people will prefer AI assistants that are stylized or assistants that are hyper-realistic will say a lot about us.
You know, I know this steak doesn't exist. I know that when I put it in my mouth, the Matrix is telling my brain that it is juicy and delicious. After nine years, you know what I realize?
Ignorance is bliss.
To be honest, as an introvert I would gladly splurge on money to have an AI "friend" to hang around with. I did something similar when chatgpt first showed up, using it as a kind of therapist buddy I could talk to no matter what time of day it was.
@@RSpwn1 I can see the Reddit mod saying this to himself as he shuts his bedroom door for a long playthrough of “Life: VR edition”
Can now envision a remote worker hired who uses an AI face 100% of the time, from interview, to all interaction, before a reveal a year later!
The “unexpected service error occurred” in the Signal To Noise section is just perfect ahahahah
Not seeing that my end, all working smoothly! Could be your end?
@aiexplained-official Ah I meant in the video @ 10:22 under the analysis of your speech 😊
The Ezra interview was wild!
High SNR perfectly captures the value of this channel.
Thank you Matt, very kind
Surely Apple must be thinking about how to implement this into the Vision Pro. For the personas.
Considering that at the end of "Her" the OS's start running on a non-physical substrate before leaving humanity forever, I don't think that'll happen in 2025.
4:33 thanks I almost spat out my tea lmao
Brilliant as always, the newsletter is fantastic too. I love the DICE scores! Excited to see what happens when you combine SmartGPT 2.0 and Llama 3
what is smartGPT ?
@@user-fr2jc8xb9g
I would call it a framework of prompting LLMs to improve their answers. It was made by the author of this channel, search for a video here.
Are these models interpolating between frames? How is it so smooth. If theres something like a cough or sneeze in the audio, would that look natural?
I do agree with the Her prediction for next year, sometimes the right custom GPT or even Pi can already feel quite close to that
I was eagerly waiting your video
You and Dwarkesh are the best youtubers in this space. I get pumped every time you come up on my front page because I know something actually happened.
All the pieces are coming together
I think that Her is already there. There is a lot of small "businesses" that sell her. After 90% of them fail, and 1% of them actually succeeds then they will plug in bigger models into it and will capture market.
Right now the progress is so quick that market doesn't have time to adjust and build evology around llms. Once we figure out ecosystem that can be readapted to newer llms/ai we will get new paradigm.
Thanks so much mate, I jumped immediately to the comments to give thee thanks for thine service to humanity. Cheers
Just finished dwarkesh Zuck interview then this drops! What a night!
One of the uncanny valley things I noticed that VASA-1 still has is that the teeth change their size and shape constantly. Once you see it you can't unsee it.
Ai explained released a new news letter that shocked the entire industry
Haha not quite!
@@aiexplained-official DICE rating: 7
It only shocked the soyboys
I am STUNNED .
Will the newsletter also cover the topics that make it in the videos or are they excluded?
Yeah, whatever I feel needs to be said. This channel will continue of course, full steam ahead!
@@aiexplained-official Awesome, I was thinking about an AI that summarizes newsletters on various topics to give you a personalized morning news briefing, kind of like your own news show. I'm glad I now have the perfekt source for the AI-News part :)
Keep up the excellent work!
No mention of the AI Index report? Or maybe I missed it?
Congratulations to 250k🎉🎉🎉
Thank you Kledda! You are the number 1 subscriber!
Always love the digestable path of the video and your explaination. Best AI channel
Thank you Rain!
Ok the skibidi toilet guy is now designing robots for boston dynamics cool cool
I would love chapters in your videos as I'm more interested in certain topics and don't have a lot of time.
The hardest and most complex ai modelling is definitely a romantic partner ai model. If they could do that right then they can do everything
That shouldn't be hard. Humans anthropomorphize everything and are emotional creatures that have a high aptitude for "falling in love" easily given strong emotional attention. There are humans alive right now that have married computers, virtual assistants and fictional characters.
I predict, that polygamy with artificial characters will be made legal to keep birthrates up.
I'm thinking it's one of the easier tasks.
@@OperationDarkside LMAO 🤣🤣🤣
Obviously not. As others have pointed out, AIs are already romantic partners for some people. Where is the LLM that correctly answers questions like "What is the chemical formula for a low-cost room temperature superconductor?" or "What is the mRNA sequence for a general-purpose cancer-killing vaccine?"
Do Patreon subscribers get a premium paid email newsletter included or is this a separate charge?
They already get all the videos, SGPT2, and much more!
There were several new papers announced recently on the power of AI to influence someone's beliefs or change their mind substantially more so than conversation with a human.
I would imagine lifelike avatars are going to enhance that even further. They will be able to train the models for optimal influence over behaviors using expressive emotion etc.
We are not prepared at all for the issues this is going to bring to our world.
I always wonder, who in their right mind thought that research into deepfakes was a good idea ? Whatever reasons they tried to justify their work, they deluded themselves.
Humans' intelligence surpasses their wisdom. Proven time and again.
I subscribe to your channel because it has no noise. I just subscribed to your newsletter as well. About this video. I've given several public talks on AI in the last year. Next month I will be giving a class to seniors on fraud, with an emphasis on AI assisted fraud. I shudder at the potential effectiveness of fraud from an AGI with voice and face mimicry.
Thanks so much leather, for subscribing to both. Let us know how the talks go, they are much needed I'm afraid.
I've always wanted paintings to sing to me. 😅
On a more serious note, you could bring more virtual people to life with this as well, which is cool.
Fantastic channel! Honest, serious attitude to accompany quality reporting & smart takes
As I had almost a full minute of spare time today with no real AI news, I did, I must confess, transgress into thinking "what if this is it; what if AI had reached its limits after all? Fantastic as it currently is, will we not see Star Trek level assistants and droids in our lifetime?" And I'm sorry, robot overlords, I ever doubted you. Please accept my most sincere apologies! 🥺
Haha
I hope it all continues at this break neck speed. I want to live in a Start Trek/ Star Wars sci-fi type world in my current lifetime.
@@vectoralphaAIWatch out coz we can get a more realistic version of the W40k😂
I’m interested in identifying as many applications as possible for VASA-1:
1. Therapy. Have it operate from a LLM specifically trained on professional therapy sessions and the DSM-5
2. Language learning. Instant feedback like you’re chatting with a native speaker. Hijacking the benefit of immersion right to your phone.
3. Speaking with celebrities/historical figures. Most celebrities/well known historical figures have many hours worth of interviews, recordings, books, dialogue, etc so train the model on those and then you can have fake convos.
4. Career coach. Load in dozens, or even hundreds of books & pdfs related to your career path, your transcripts, and resume and have a fake career coach.
5. Apple Vision Pro/VR - I just know there’s something here.
What else? And I understand there will be tons of nefarious uses but I choose to also try and think about positives. Also, I know most of these are happening with GPTs but not like a live conversation with a realistic persona.
Language learning and Career coach are the most anticipated for my part , the best use i currently have for AI is learning , its substaincially better than anything else at least in beginner level topics , because its much faster than searching online or reading books , and more available than a professor , what vasa could do is make this process even faster because speaking and listening is faster than writing and reading , although rading is better for comprehension but speaking to the avatar is objectively better than prompting...
Hey, I am really enjoying your videos, and like your content. I also singed up for your newsletter, I have one question, isn't it normal to get an email at the time you sing up, to make sure its really your mail, and that you have to klick a link on. Otherwise, someone could just sing up many different emails, and it will be like scam for them and you might get troubles.
But I am not sure about the laws in America, but at least here in Europe it must be like that.
Maybe consider this :)
Great idea! Will bring one in! Thanks Hans
10:15 regarding the calmness .. I think hume has one bucket for that emotion. Having a low calmness score means you were quite pumped. Perhaps they should have pumped you are category was well?
I tried for an annual subscription. There were no errors from the e-commerce links; there were no field labels either. No button was enabled.
Those realistic avatars with voice cloning are going to create havoc when spearphishers get their hands on them.
Thank you for sharing your time and work Phillip, have a great day brother, peace
Thank you Bill
@@aiexplained-official You're welcome Phillip, after watching a few interactions with Yann, it seems like he's doing everything to create his own reality in the hopes that people will buy in to his predictions, I'm still trying to figure out the purpose of his wet blanket routine, but its getting old.
Vasa is interesting. Wish you could type a sentence and the model is able to use the sample audio clip and the text to talk. I am sure it has its downsides as well. Imagine someone who has speech impairment is now able to use this to converse.
6:40 Honestly, that's still on the uncanny valley for me, so yeah, I can imagine and it would be very uncomfortable. For now.
Looks real to me.
Yeah the side movements are uncanny but i could get used to it
Yeah, not quite there yet.
The eyes are sometimes wide open, and when you add the sideways movements they make, it looks uncanny
I think it's right on the edge of uncanny valley, if you're unaware it's generated video, there's a good chance a reason person will believe it's real.
Now that we have real-time avatars with natural-looking emotional expressions, we need AI models with personality and innate emotions to use them properly.
I have to admit, even though all modalities are interesting, the only thing I am really interested in, is the ability of these models to plan and to reason.
The rest is kind of 'normal' technological progress.
I know exactly what you mean, that is my central focus too
Movie "Her"
AGI
Signal To Noise newsletter
Good moments:
2:12
5:20
6:38
9:00 Vasa-1 #darkweb #opensource
10:00 HUME AI
As always thank you for your insights...
Whenever a expert gives some number of years before AGI happens, I cut that time in half. It will take us by surprise.
One thing I would like you to talk about is the adjacent technologies that are going to get pushed forward due to the overwhelming demand of AI such as the need for much greater internet bandwidth etc. Which areas do you think would stand to benefit the most from these developments and why?
GPU, etc. development obviously
I agree that we probably will have Her level AI around 2025.
I am so glad that I found this channel. Has been great to learn and keep myself updated.
The catastrophic risk discussion about AI, starting 15:36 is about an important topic
Excellent video as always
Note gpt4 turbo is multimodal not text-only like Llama3. That’s a big deal to not compare them side to side.