[ML News] OpenAI is in hot waters (GPT-4o, Ilya Leaving, Scarlett Johansson legal action)
Vložit
- čas přidán 20. 05. 2024
- #gpt4o #sky #scarlettjohansson
After the release of their flagship model GPT-4o, OpenAI finds itself in multiple controversies and an exodus of senior personnel - notably Ilya Sutskever
References:
openai.com/index/gpt-4o-and-m...
openai.com/index/hello-gpt-4o/
x.com/LiamFedus/status/179006...
x.com/lmsysorg/status/1790097...
x.com/bindureddy/status/17901...
openai.com/index/improvements...
openai.com/index/openai-and-r...
archive.ph/jHlMm
www.vox.com/future-perfect/20...
x.com/soumithchintala/status/...
x.com/sama/status/17919368575...
x.com/ilyasut/status/17905174...
x.com/sama/status/17905180316...
x.com/janleike/status/1791498...
x.com/sama/status/17915432640...
x.com/gdb/status/179186913813...
openai.com/index/how-the-voic...
www.forbes.com/sites/robertha...
x.com/BobbyAllyn/status/17926...
x.com/stclairashley/status/17...
Links:
Homepage: ykilcher.com
Merch: ykilcher.com/merch
CZcams: / yannickilcher
Twitter: / ykilcher
Discord: ykilcher.com/discord
LinkedIn: / ykilcher
If you want to support me, the best thing to do is to share out the content :)
If you want to support me financially (completely optional and voluntary, but a lot of people have asked for this):
SubscribeStar: www.subscribestar.com/yannick...
Patreon: / yannickilcher
Bitcoin (BTC): bc1q49lsw3q325tr58ygf8sudx2dqfguclvngvy2cq
Ethereum (ETH): 0x7ad3513E3B8f66799f507Aa7874b1B0eBC7F85e2
Litecoin (LTC): LQW2TRyKYetVC8WjFkhpPhtpbDM4Vw7r9m
Monero (XMR): 4ACL8AGrEo5hAir8A9CeVrW8pEauWvnp1WnSDZxW7tziCDLhZAGsgzhRQABDnFy8yuM9fWJDviJPHKRjV4FWt19CJZN9D4n - Věda a technologie
I have to say I just don't really hear the similarity
Exactly! There is a similarity with the general tone of "Samantha" in "Her", as it talks in the same style, but definitely NOT the same voice.
I don’t think anybody does. Most people seem to be going off of what others say, rather than listening themselves.
@@float32 A timeless problem among human beings :P
iirc some ppl criticized how scarlet johanson got greedy with disney a couple years back. this could be a confirmation of her character
it sounds like Rashida Jones tbh
Don't worry about misaligned AI, do worry about misaligned humans. There is really a lot of precedence for misaligned humans.
I would be more surprised if you could find a decent number of humans aligned to the good of humanity lmao.
Exactly, if AI goes bad it'll be well aligned AI with misaligned company owners. The same owners of that have shown no shyness to employ behavioural psychologists to exploit our psychology whilst using their apps will soon have super human capability to accurately predict our response tokens from any given input they show us. If I can predict with accuracy what you'll do in response to what I'll do, I can manipulate you any way I want. This will give them super human manipulation capabilities and what do we think they'll do with that considering the history of these same companies? They wont pause for a second, they'll jump in it
Of course, yeah, some humans and corporations are currently far more 'misaligned' than the AIs are... 😉 The LLMs are just 'mirrors' reflecting back society, no will of their own, whereas the humans... can be actively malevolent.
of course, but that is what alignment research is all about. you really don't want these things making biological weapons research available to everybody at a high school level. So I think that Yannic is being *slightly* naive here.
@@rickymort135this is not logical. Being able to predict what somebody will do does not imply that you can control what they do. In no way does that follow logically.
Yeah, sorry Yannic, but Scarlett's voice is just a normal female voice. "Sounding identically like" Scarlett is 100% legal as long as her name and image is not mentioned in the product. The only reason why Sky as been removed for a while it's because it's probably instructed that it is an female AI. and It saw the movie Her in the database. But there is no legal case there.
That Scarlett Johansson controversy is hella sus
There are safety concerns related to prompt injection and other newer methods for LLMs hacking. Letting LLMs access confidential data through various upcoming integrations seems risky. But yeah super alignment issues seem very far from the current stage of LLMs.
Informative video, Thanks!
Yeah LLMs are inherently insecure. They should never be in charge of sensitive data without additional security. But that’s very far off from what these people are saying about how AI is gonna cause world collapse and like the extinction of the human race haha just seems a mega dramatic
Why would they even want Scarlett Johansson? They should have asked GLaDOS' voice actor!
Cake!
Amazing video thanks for always keeping us up to date with all the developments in this moving field
I understand the lawyers of Scarlett Johansson. The bigger area in voice similarity space they claim as copyrighted, the more money for Her.
I wonder if they took it down to avoid any secret sauce that might be exposed through discovery?
It wouldn't be a copyright suit, it would be a voice misappropriation suit. They would essentially be suing under the claim that the average person would come to believe that Johansson endorsed OAI through their voices sounding similar. In the case of impersonation, voice misappropriation can only apply when a person's voice is distinct and a central part of their identity; dubious claims when it comes to Johansson. Historically, this has exclusively applied to professional singers. It's also questionable if the voice even qualifies as impersonation to begin with, but, ya, the suit would likely not end in Johansson's favor (unless the parties settle).
Nice
considering his tweet, ilya is serious about keeping his shares
Of course he is, you realize how much money that is?
possibly also about his personal safety
Imagine you wanted to avoid dangerous levels of greenhouse gas emissions today and had access to a time machine. Surely, you would want to travel back to a time as early as possible to maybe the first prototypes of steam engines and hand over the blue prints for clean air filters and push for everyone to implement them from the beginning. But if you explained to the people in past why this is necessary they wouldn’t believe you because they couldn’t imagine that these early prototypes eventually become so widespread as to cause global issues. And they would be kinda right to distrust you, since their early prototypes were so primitive that none caused major problems.
But where else would you intervene except as early as possible? As soon as steam engines are widespread enough to cause harm, changing them becomes a lot harder and people will oppose the associated costs.
That’s the situation that we are in with AI except there is no one from the future to tell us what we should do to avoid danger. We have to figure it out ourselves ASAP while the world pushes forward with rapid adoption. That’s why we need safety research and regulatory oversight with regard to capabilities.
(Btw you don’t have to consider climate change an existential risk to understand the analogy I am making)
If my grandmother would have had wheels she would have been a chariot.
The more I learn about Revux, the more I believe in its vision.
29:02 - did you use the word "brabble"? I've never heard that word before, but looking up the definition it definitely sounds like it fits.
How dare people to have similar voices to famous actors?! Now they have to wipe their tears with millions of dollars Q_Q
If the training wasn’t from an actual voice actor, there would be merit. Scarlett Johansson doesn’t own other voices that sound like hers.
I believe that Sam Altman really made a big mistake with the "her" tweet and the prior and recurring contact towards Skylett (pun intended, probably on OAI side too) Johansson... It sounds a lot like an unauthorized use of identity, and has been successfully litigated by singers and actors over time. SJ is one of the few to have successfully won/litigated against Disney... so I wouldn't be so sure of the "no merit" here.
They have to prove the voice they used isn´t sampled from the movie Her. Clearly they cannot, otherwise they wouldn't have been asking her repeatedly and finally removing the voice.
@@VangusKhan "stolen" and "trained on" seem like two words with significant distance between them... Are you sure you want to use them together?
I'm not a lawyer but this reminds me of Ellie from The Last of Us initially being much more similar to then Ellen Page. Naughty Dog finally caved and changed the model. They didn't use a scan of the actor's face but they probably asked an artist to steal her likeness. Clearly OpenAI asked a voice artist to imitate Her as closely as possible. Maybe it's illegal, maybe it isn't. But it's an optics nightmare either way and OpenAI's statement doesn't even try to convince people otherwise. They themselves say AI shouldn't imitate celebrities. It's obvious they did.
@@FrancoisPesce How dare he post a pronoum on Twitter!
I heard your opinion, here's mine. Superintelligence may be far ahead, it's not about probability of it happening but risks.
It makes no sense to wait for it to happen to start trying to solve or regulate the problem.
Were it to happen, risk is huge and reactivity cannot be any effective without proactivity.
Idk whether it's gonna happen or not? Neither when?
But the tiny chance of it happening is enough to care about it and if it happens we will be very grateful to those people who cared to anticipate.
Profits and short short term utility are driving too much our societies' trajectories.
In short, the worst-case scenario is always worthy work and waiting for it to happen is always too late!
Well, superintelligence/alignment might be a real thing, but probably not for 10 years at least, and corporations might not be the best place for it to come from.
Regulation almost never helps the consumer. Regulation 100% of the time helps the incumbent corporations. Guess who would write the AI regulations and has been advising congress on it… yeah it’s OpenAI and Google obviously. Regulation is 100% about protecting large companies and not at all about protecting populations
The sky may fall on our head. Asteroids may crash into the Earth and eliminate human civilization. Supervolcanoes may erupt. New deadly viruses may be engineered by tax-funded military labs in USA and China. What are we proactively doing to prevent those disasters (especially the last one)?
I see Revux as a long-term hold. Great fundamentals!
Honestly, the voice delivery & tone , sounds a lot like the character in “her” specifically, which doesn’t sound like Johansson’s speaking voice in other roles * interviews (but I’ve no idea what she sounds like in day to day conversation).
I would’ve expected the studio that owns that movie to be the first one to take legal action (given how copyright-happy the studios usually are), but I guess the fact people immediately started associating the voice with Johansson & the fact Altman had been trying to get her for the voice, is what prompted her to take action.
But voice actors can and do imitate many voices, especially of celebrities and famous people. They are not sued for that. It is arguable whether you legally "own" your voice in the sense that you can prevent anyone else speaking like you. Even if we ignore the fact that given billions of people many of by necessity have similar voices. Still, chosing to imitate her voice after her explicitly refusing is a certified a-hole powertripping move.
Steve Perry left the band Journey so Journey found a new lead singer who can sing pretty much exactly like Steve Perry... no one lost their sh*t. This is sort of normal. If they wanted that effect, they have a right to find a way to achieve that effect as long as they don't actually steal her voice.
Revux is making moves quietly. Keep an eye on it!
Looking for where to sign that I know that the voice is not S.J. so we can all move on with our lives and just get the voice I've gotten used to from openAI back up and going...
OpenAI really sounds like a harsh 'cut-throat' corporate environment, they do make some cool products though. The drama seems to have blown up into a bizarre political fight, it's a shame really - everyone piling on to OpenAI. I get the feeling we'll never get the full story of these happenings for years or even decades ...
If you're looking for a solid x100, Revux should be on your radar.
Sam Altman saying that he pinky promise didn't know is the least believable thing I've heard all week
Lying every day, all day long, to a lot of people is kind of part of his job.
I am suspicious of Sam after that firing drama. He seems a charismatic guy with hidden motives...
Just shifted my focus to presales, especially Revux. Early investment is key!
GPT-4o image generator is transformer? rather than diffusion?
it's one model, so yeah, it's a transformer
Isn't it a combination of architectures?
Dropped all my small caps - focusing on BTC, Revux, and maybe some ETH.
I can't get a simple for loop to create some sql statements for me generated correctly after 4 correction attempts *with parts of the correct code* - we've a little way to go, when taking into account 3 contexts simultaneously (the generator itself, the sql, and things in the sql) looks to be impossible ...
that openai agreement is insane - gotta be crazy to sign that
i would add two things though
~Scarlett was asked in September
~The voices were released in September
hence the prep work, casting, selection, recording were done prior
I really liked this video's takes, until we got to the Skye situation. In the OpenAI blog post which you referred to in the video, they themselves stated that they were cast and recorded the ChatGPT voices over the summer, which lines up with them being released to ChatGPT Plus subscribers in September. So when they reached out to Scarlett, the Skye voice had already been created. They could be lying about their timelines, but I'm inclined to believe them because I, and many other plus users, first heard Skye's voice in September when the feature first rolled out to us.
You seem to have very interesting / insightful videos as a whole, so I'm a little surprised that you missed this detail, especially since you did a pretty good job of covering both the optimistic and cynical takes.
Chatgpt-omh we didn't make it to AGI ..
so they hired a voice actor that sounds similar? it's not like that's never happened before.. it happens in commercials all the time. As long as they aren't calming it's actually her.. but, that tweet was a little misleading in that regard.
Correct me if I am wrong, but I remember in the reaction to the original Gemini demo six months ago which claimed that it could parse video in real time and generate a response. You (correctly, possibly) were claiming that of course it was faked and nobody knowledgeable enough would fall for that.
So it is worth pointing out that either this point of view was completely wrong six months ago, or the progress since then is kinda insane, right? And I think it's worth finding the answer to this.
I thought the Gemini video was doable, but the dedicated bandwidth and server resources to transfer, ingest, and process a constant video stream would cost... $10 a minute? Has OpenAI announced pricing for real-time video?
@@skierpage Check out JETSON AI LAB | Realtime Video Vision/Language Model with VILA1.5-3b and Jetson Orin
Shopify partnership puts Revux on the map. Big boom incoming!
it's lame what they've done, but "sound alikes" are legal. long history of them, especially in music. it has to be different enough, but, legal. It's immoral and not right, but, lots of things are legal and "not right."
regardless, ScarJo is doing the right thing by making it publicly known that she objects to her "Her-like voice" being used in a Her-like AI product that she hasn't agreed to.
But the Sky voice doesn't really sound like her in "Her"! It's bubbly and chirpy, not breathy, deep, and sultry.
The Sky actress is protected by the same laws that Johanson is. She has her own voice, applied for a job that Johanson declined , got the job, did the job, and it is her IP. Not Johansons IP.
@@Ben_D. Your voice isn't IP. I ANAL, but if OpenAI hired her and told her to imitate Scarlett Johansson in "Her", ScarJo has a case. Even if OpenAi just hired the voice actor with a speaking voice close to ScarJo, there still may be a case of misappropriating her voice, violating her "right to publicity". I'm torn because to me Sky _doesn't_ sound much like ScarJo in "Her", but reasonable people strongly disagree. Read the Verge's article "Lawyers say OpenAI could be in real trouble with Scarlett Johansson" for some legit-sounding legal analysis.
@@skierpage Scarlets voice, image, name, and so on, ARE her Intellectual Property. None of those things were applied to Sky.
I know that OAI backed down, but I doubt it was because they feared losing. They just dont want more battles, even if they would win them. Winning would probably be a Pyrrhic Victory.
@@Ben_D. it's unfortunate you're doubling down without bothering to read the short article I provided, or WIRED's article on the legal aspects. No, your appearance and voice are not intellectual property! Which is good otherwise the word "intellectual" in IP would be meaningless.
Not a shill, but that Times Square announcement was hard to ignore. Cyberopolis will be big!
Hmmm... if you follow Evan Hubinger's of Anthropic new series about sleeper agents and their research about how to handle something like that you must come to the point where you have to aggree that every company seriously engaged in developing models should have good AI safety researchers depending on the usage of the models.
Not to be aligned to humans when models hallucinate or do some unwanted NSFW stuff is one thing, but waht about the mass of companies which specialize in developing models for military usage or controlling of critical infrastructure?
Wat contracts do we know about OpenAI has made hidden to the public for other use cases then chatbots or robots?
Imho. It is a bad sign that these excellent engineers left the company for the reasons they published.
Hey, the military is for "protecting" us, did you forget? Do you think your silly AI safety researchers can stop bloodthirsty people with guns (or drones)? Have they ever?
Of course it's a bad sign.
The principles went out the window when they tasted the money...
Enjoyed the commentary even though I knew all the topics reported quite well.
The GPT-2 too dangerous quip is getting a bit old though.
Keep it up it's entertaining and educational.
Can we imagine the future?
Thank you Yannik for explaining the recent drama at OpenAI in this episode of ML News
18:42
"You know what's a good curve?
Track the performance of computers vs humans in long integer division or playing chess or something like this.
That's the situation we find ourselves in except the axis is like statistically modeling language or something like this.
Sure you can do more stuff with it but it is wildly different from super intelligence and dangerous super alignment."
I feel like I’m having to do some serious mental gymnastics to try to understand what Yannic is actually trying to say, especially the “Sure you can do more stuff with it” part.
Does he mean that there’s still a lot of scaling needed to get to any significant or general-purpose capabilities?
Or does he imply that modeling a set of data can never lead to any significant or general-purpose capabilities?
The end result of building a bigger computer and/or using better algorithms to perform better long integer division is a system that does better long integer division.
The end result of creating a new algorithm or training a bigger model to play chess is something that plays better chess.
The end result of training a larger model, such as a typical transformer based LLM, on a larger dataset is it learns to better model the data. Literally any digitized data.
How are obviously narrow tasks like long division and chess analogous to modeling any dataset? I understand the "wildly different from super intelligence" part, but trying to compare LLMs with long division just sounds incredibly silly.
agreed
Yannic is great at technical matters and sarcastic takes :) However, AI safety is his blindspot. He seems to completely ignore it.
When I subscribed to this channel it had one of the best selection of relevant AI papers paired with their analysis outlining the important thoughts. So it was a cool way to stay ahead of some broader research trends. Now it is becoming one channel more for AI related business news. This certainly mirrors an unhealthy development which Yannik has repeatedly mentioned: AI moved in the last 18 months from an open research field to a business field with competing companies using tech information mainly as marketing tool. But the intellectual emptiness of all this is tiresome. It would be great to get more scientific content again. And as long as there is no way around the commercial llms, maybe reports on rigorous tests and similar research is a good counterweight against the stream of marketing hysteria. I seriously miss the old channel.
Pausing is the right thing to do. But it is not an admission of guilt.
The tweet shows the inspiration from the role of that movie, but not the actress.
I must that movie is the first thing that comes to mind when you see the demo. But listen to both voices, they don't sound the same at all. What were they suppose to do when scarjo refused, cancel the role ? Our AI is not gonna be flirty because the actress was flirty in that movie ?
I'm pretty sure the "her" tweet was just commentary on the live interaction that they were demoing the same week. That demo used many voices, including the sky voice, and the sky voice had been released several months before the tweet. It doesn't really make sense to interpret the tweet as being a reference to the sky voice.
only what you can measure you can also optimize. It's clear that when there is an arena as the measuring stick for AI - they are going to try to optimize for that.
Non-disparagement clauses tell you everything you need to know about the company. It's like wanting a prenuptial agreement before you'll marry someone. If you feel you need one for the person that you're supposed to be spending the rest of your life with then you probably shouldn't be marrying them in the first place.
I miss Sky. About alignement, i think there should be research on how to control and align ongoing in parallel but agree we are nowhere near having to worry about implementing it. And it shouldn’t be a blocker in development or deployment. Bad actors will figure out how to get ai powered weapons. There is no stopping that. And that will happen waaay before a rogue AI is possible.
OpenAI is giving the same vibes as the 'downfall of the walking dead' doc I watched the other day. They really need to make more decisions based on optics if they want to keep the goodwill they have found themselves the stewards of, which is a strange thing to say and not normally the case. Maybe the EA philosophy is the culprit? I can see them being successful, even if they stay on this track that makes them look like creeps, but it is going to be sad if we have to watch them grandiosely release a series of "AGI" products that aren't (like 'full self driving' is to tesla).
very based takes 👍👍
I share your opinion about AI. I don't think in any way AI is a danger right now. It is progressing fast. It has a very huge potential. And open source should be the way. As you mentioned. Llamas are open and no one destroyed the world and even if it was 1000 times more capable, still, no one would destroy the worlds because the laws of physics are clear. To destroy the world you need resources and it is not just intelligence.
I really do not agree that we are far and safe from seing dangerous super AI. And seeing researchers leave should makes us concerned.
The real problem is that we do not know how far we are. Solving alignement is very difficult and it makes sense to work on it as early as possible.
Not here to shill, but the Times Square ad for Cyberopolis made me Google it.
Just a note: Speed **is** performance on benchmarks. Well, I mean, not literally, and not when doing them as conducted, but LLMs can execute strong performance at inference by generating more responses in a variety of frameworks, so I'd say while it's definitely exciting to get new models with stronger zero shot capabilities, speed isn't exactly a bad metric to see solid performance in.
Just use a wake word to have assistant features in a REPL. I do that in SimpleAgent.
What about GPT5?
I still have the Sky voice and I don't think it sounds like Scarlett. I'm confused.
Yeah. Hogwash. That's what these concerns about superintelligence are. At least for now
All I want is to just tell my lawn mower to mow the lawn
LUBA 2 AWD 5000: Perimeter Wire Free Robot Lawn Mower, only 3000 Euro.
SB-1047 Coverage please
It is not about aligning the GPT, it is all about aligning the users to a certain perspective of the world. My 2c.
Finally someone that gets it!
I think the voice problem isn’t really the voice it’s more that he explicitly called out the movie in his comms. I worked for a company that used a Hollywood movie name as the internal project name for a PC motherboard. This name leaked and the name was bandied around the internet. We were sued by the studio and lost. We also hit sued by Frank Zappas estate for a very similar reason. This is why companies have legal😊
4o is not targeted to be the new 5. It is the new 3.5
my prediction: Ilya will be working for some low key military/arms stuff (maybe Jan too since hes so worried about (super)alignment) **cough
Qventi is everywhere now, just like Revux was.
Bring back sky!
Caprice de star 😎
Yup...
Love BTC the most, but it is time to put some into presales. Cyberopolis is the next big thing.
On the other hand, if there really was an actress with similar voice, not considering her just to create distance from Scarlett Johansson would not be fair. There are 2 sides to this coin.
When listening to the GPT4o voice demos, I thought it was kind of over the top, exaggerating emotions, with a kind of California influence although not Valley Girl.
At no point I thought of Scarlet Johansson tho.
It sounded kind of low sample frequency also, not the highest audio quality.
I don't think LMSys gives Claude as an opponent much in the chat arena.
seeing openai going into the dark end for itself feels weird
8:55 BOTTOM-UP-AI ... Alicization arc confirmed!
when I saw the demo, my first thought was "sounds a lot like scarlet". what is messed up is Sam is obsessed with "her" and decided to go ahead and copy her voice. I think of it this way. If it was my daughter, how would I feel about a creepy AF tech bro making a voice like my daughter. Sure he isn't going to play some creepy voice fantasy with this voice.
when scarlet rejected Openai, they should have stayed clear and just chosen something unique.
you don't get 700 out of 750 people petitioning your reinstatement to OpenAI without some
financial gratification
For the benefit of all humanity (*)
(*) except Scarlet Johansson
I just love this channel
18:23 we in principle can falsify that statement by actually building strongly superhuman AI and it not ending up very badly (without exerting specific effort for that I mean)
It is funny you say that openai do whatever "misinformation" and "deep fakes" they desire "suddenly" as an anecdote immediately after the story of the guy who cared about the safety issues leaves
Bro is back 🗣🫡🔥
Talking robot toys this Xmas
ScarJo has apparently trademarked the female voice. All other women should keep quiet.
Honestly I'd fully support OpenAI in this, if it weren't for their hypocrisy, as you said. The use of another actress' voice, and deliberately going for the concept of a flirtatious AI assistant like Samantha in the movie "Her" sounds like it should be, by all means, fair use. It makes no sense that they hold the rights to a sci-fi idea, and that they can veto its eventual practical implementation.
However, their holier-than-thou attitude has been extremely annoying. You reap what you sow. If that's the game you want to play, you had it coming your way, and people won't be bothered to defend your asses.
Better call Saul
Wait... are you trying to tell me we can't trust the guy who duped 2 million people in to trading their biometrics for an altcoin?
How can you not trust these pretty Manson lamps?
OpenAI really blew it by choosing generic hollywood actress #827312 instead of cute anime maid
Even though the Anime Maid fanbase is used to spend a ton of money on anything similar to their waifu. Totally wasted opportunity.
Fuck, imagine how much dough they would be swimming in if they had hired Tara Strong!
More like "Hello I'm Sandy I'll be your server today!" generic bubbly California waitress. Scarlett Johansson's voice would be an improvement.
What if the tone is like in the movie Her? Then you can't do it in real life? So we can't build space ships because Star Wars would sue? Or flyng skate boards because Back to the future would sue? Also why is SJ the one suing? Does she have the rights to the AI character in Her? I get that she woule want to protect her voice, but the idea that she would gate keep a concept, enthusiastic/flirty AI, is ridiculous. By the way I don't think flirty AI is a good anyways so this is not me stanning openAi 😅
Scarlett sounded like Siri sounded like Majel. Choosing to have a female voice with emotion is the least interesting thing about open AI.
Imho it's not. - Having a model that may adjust it's voice's mood in reply of a scan of the mood the user is in will be superior for commercially used chatbots like service-chatbots.
And because you can't trust any LLM's output at the moment when it comes to data, chatbot's are, truely said, the almost only serious application for LLMs at that moment in time.All other AI applications can't be trusted at all at the moment, every data has to be counter-checked after the processing by a LLM.
So to have at least one really good functioning AI application it's a big gain when that get's the feature of adjusting the voice to the actual application situation.
@@henrischomacker6097 I agree that their implementation is pretty amazing, but the fact they tried a personable woman is completely unremarkable.
It's Johanna Scarletsson 😑
Perfect waifu is gone? What am I going to do without perfect waifu? WAIFUUUUU!
The Sky has fallen! The Sky has fallen!
Even of Scarlett doesn't have a license to similar voices, I think deliberately choosing a simialr voice to make an impression of the character she played is very much an issue, even if not necessarily a legal one, becuase voice is the most important part of that particular character, so they are essentially plagiarizing a character, rather than a voice.
But the Sky GPT-4o voice doesn't sound much like Samantha / Scarlett Johansson! It's bubbly and chirpy California waitress, not her deeper, breathier, more sultry voice. And you can't claim copyright on a helpful interactive AI companion.
@@skierpage Well, normally you wouldn't be able to, as long as it's not *too* similar. But given the history of OpenAIs interaction with Scarlett, it still sounds like a tricky situation, though I haven't heard the comparison between the two myself.
And I'm pretty sure if the AI companion tried to copy Batman, especially after being refused an official partnership with DC, it would absolutely be hit with a copyright claim.
The question is whether there was an intent to make it similar to the original voice, but different enough to make sure it's still legal, or if it's simply a coincidence. If it's the former, even if it's not necessarily an issue from a legal standpoint, it's still a nasty move nevertheless. If it's the latter, then it's just bad management on Sam's part, since he didn't think of the implications of releasing a voice that sounds similar to Scarlett after being rejected. Even if the court throws the lawsuit out, it's just being confrontational for no reason. There are thousands of different voices, why pick one that has potential of causing problems?
@@skierpage Just listened to a few comparisons, and the voice actually does sound quite similar, so I think the question of how it was generated and whether the similarity is intentional is a valid one.
Revux's use case is exactly what the crypto world needs.
This was supposed to be GPT5 but relabeled as 4o due to its inferior performance. The reason I say this is that multi modality and tokenization revolutions mean training from scratch, and that is too expensive for just an incremental release. This was supposed to be GPT5, but such a disappointment after a year and a half from the previous model, would have been shocking for investors, so it was relabeled as something "incremental".
I think it is a shame they removed Sky voice. Maybe if they offer big bucks to Scarlett we can have it back.
It underlines one thing: it is OpenAI which decides what you get or not, it is not you, OpenAI's paying customer. Same story for Microsoft. And any striving monopolist. That alone is a sufficient reason to avoid all their products.
@@clray123 I think their product is pretty good and I will use it for as long as there is no alternative. Llama3 is in the ballpark, some times, it gives better answers that chagpt.
At least it cannot smell my room after some chilli con carne.
tbh OpenAI sounds more like China every day.
What ai doomed is describing is basically what if we created a Omni potent ai god 😂
On one hand, it makes sense to save money by avoiding the legal costs and just throwing away an innocent voice; on the other hand, John Hammond spared no expenses....
I've only used the web version but I can say for certain that it can't see jack. I uploaded a floor plan image and asked it to colorize it. To say that it failed would be generous. If it can't convert the image input to text in a meaningful way it sees nothing. I also gave it a frame from the duckman cartoon and asked it to recreate the scene and it was clear that all it could see what a generic text description of the input image and a text prompt was also the input for the image generator. There was a vast logic gap in the image to image process. On top of the monumental failure, GPT4o proudly declared detailed success on all points.
I think you might be misunderstanding the model, GPT-4o does not output images at all. Any time it is outputting an image, GPT-4o is writing a prompt to DALLE-3 in text and then it generates the image. If you’re passing it a floor plan, of course it’s not going to give DALLE a good prompt, because imagine trying to perfectly describe a floor plan through a constrained text prompt. It can, however, take in images as input, and I’m sure if you asked questions about an image it could answer them decently
I’m pretty sure GPT-4o does not output images at all. Any time it is outputting an image, GPT-4o is writing a prompt to DALLE-3 in text and then DALLE generates the image. If you’re passing it a floor plan, of course it’s not going to give DALLE a good prompt, because imagine trying to perfectly describe a floor plan through a constrained text prompt. It can, however, take in images as input, and I’m sure if you asked questions about an image it could answer them decently.
gpt4o does not have its image output enabled for the public.
What you are seeing is just Dalle3 via gpt4o.
I think these AI companies are shooting themselves in the foot with these half-assed "products" they release. But by now they must be in desperate/panic mode as they realize that their marketing hype/scam is becoming unmasked every day and that the big investors start getting sour about having been (voluntarily) conned into jumping on this bandwagon which is about to crash.