The world may be about to change! OpenAI launched GPT-4o on Monday, which understands what the camera sees, and you know what that means? This is equivalent to AI having eyes from now on, and can communicate with humans in real time, without any delay, and the tone is basically the same as that of humans.
I tested it myself, and even when humans were talking and thinking, the pauses and repetitions were all accepted as they were, and there was no sense of disobedience. This means that there is really a robot talking to you in the future, you may not even know that it is a robot, if it has the appearance of a human again, is it completely unimaginable, science fiction is really about to come true.
This live video released by OpenAI was hosted by OpenAI's chief technology officer, Mira Murati, and demonstrated by two R&D leaders, Mark Chen and Barrett Zoph.
When Mark used the app on his iPhone to evoke ChatGPT, saying that he was a little nervous about the live demo right now, and that he could help me, ChatGPT replied that you just need to take a deep breath and remember that you are the expert here. Isn't it a little humorous?
Mark deliberately gulps in and asks ChatGPT to give feedback.
ChatGPT says, oh – slow down, you're not a vacuum cleaner.
(Wo! slow down. Mark, you are not a vacuum cleaner)
This shows that ChatGPT detected his emotions from his voice, which caused laughter. Mark said ChatGPT can also handle interruptions, change the tone of voice, and analyze the user's facial expressions, among other things.
In the subsequent demonstration, it can be seen that ChatGPT did achieve a seamless connection, smooth dialogue, and was constantly interrupted while telling a story, asking to change the tone, and did it immediately, and changed it to a song as required.
Barrett then wrote down an equation on a piece of paper, and Chat GPT helped him solve it smoothly following the steps. At this time, Barrett wrote "I♥Chat GPT" on a piece of paper and showed it to her, but Chat GPT was so moved that he almost laughed out of a pig cry, saying that you are too caring, and then praised Barrett for wearing good clothes and looking like his emotional intelligence is also a bargain.
Barrett then interacted with ChatGPT via a piece of code and a weather graph, and ChatGPT smoothly interpreted the code and images. At the request of the audience, Murati demonstrated the use of ChatGPT for real-time translation, Barrett asked ChatGPT to analyze his expression through the camera, and ChatGPT completed it without any delay, in short, it was like a group chat with several people, if you only listen to the voice, you may not feel that there is an AI talking inside.
According to OpenAI, GPT-4o can "respond to a user's audio cue in as little as 232 milliseconds, with an average of 320 milliseconds, similar to the reaction time of a human in a conversation." ”
Isn't it completely amazing, and Google's demonstration of Gemini last December seems to be similar, but Google's demonstration is considered to be exaggerated and misleading, and it is made by editing, so Gpt-4o, is there also an exaggerated and misleading component?
I have a feeling that this may be the case, because Mark and Barrett's demo, which must have been designed in advance, included Mark's loud pant, Barrett's equations, and the code. It's hard to say whether you have rehearsed with ChatGPT in advance, but at least there should be a draft, of course, ChatGPT may indeed have this ability, and many friends may have a feeling that the longer you use ChatGPT, the more routines it has, and the more familiar you will be, and the more you communicate, it will become a rambling conversation.
Then I immediately opened ChatGPT, and sure enough, it reminded me if I wanted to use Gpt-4o for a conversation, and I didn't hesitate to agree, but the surprise was that it was completely different from Murati's demo, and I had the same conversation and use text chat as it was, and it still had to react for a long time to respond, and there was no response of 232 milliseconds at all, but its voice was really different from other AI, and it sounded like a human voice.
The most incredible thing is that it also answers my questions about the response time, and there is a screen recording in the back, do you think it exposes the inside story of OpenAI? OpenAI, like Google, is it also designed? Here's a real conversation I recorded.
----------------------------
I:
Well, yes, when I'm standing in front of a very high place, like the railing of a tall building, sometimes I have a terrible thought, that is, I suddenly want to jump, what kind of phenomenon is this, has anyone really jumped like this?
----------------------------
ChatGPT：
The feeling you describe is called the call of the void or high place phenomenon. This is a fairly common psychological phenomenon in which many people suddenly feel the urge to jump when they are in high places, but this does not mean that they really want to harm themselves or have suicidal tendencies. As to why this feeling is the way, psychologists believe it may be a reaction of the brain as it processes fear and survival instincts.
As for whether or not anyone actually takes action because of this impulse, this is very rare. Most people who feel this urge quickly realize that it is just a sudden thought and will not actually take action. If the feeling is upsetting or bothering you, it may be helpful to talk to a friend or professional about your feelings. Have you ever experienced a similar situation or have other questions about this topic?
----------------------------
Do the last two sentences sound familiar? It's starting to roll around again.
----------------------------
I:
Well, thanks, but I see OpenAI says that you can respond to the user's audio prompt within 232 milliseconds, which is similar to the reaction time of a human in a conversation, why did you respond for so long? Is the model you are using really Gpt-4o?
----------------------------
ChatGPT：
Yes, I use a GPT-4 based model. Response times can theoretically be fast, but in practice they can be affected by a variety of factors, such as network latency, server load, and the complexity of the issue. My goal is to answer questions as quickly and accurately as possible, and sometimes it may take extra time to ensure the quality of the answers. If you have a question about urgency or a specific time need, just let me know and I'll try to get back to you quickly. Is there anything else you'd like to know?
----------------------------
I:
I watched your demo, it's really a second response, it's really amazing, what is the reason for this?
----------------------------
ChatGPT：
Thank you for your affirmation! Our models are trained on large amounts of data to improve responsiveness and accuracy in a variety of conversation scenarios. It performs well in the presentation, probably because the problem is relatively clear in these situations, and the model is optimized and tuned for these common problems to ensure a quick and accurate response. In actual use, the model will continue to be optimized according to different problems and uses. Do you have any other questions or areas you'd like to know about this technology?
----------------------------
Looking at these two dialogues, does it feel like this presentation was really designed in advance? Because of my actual conversation with it, it responds very slowly, not at all like the second answer in the demo. What's even more telling is that because the previous question didn't flow well, I repeated my question, and it answered basically the same as my previous one, which meant that it was completely possible to design and rehearse beforehand. In addition, I asked it to sing what it was going to say, and it also said that it couldn't sing directly, which seemed completely different from the presentation.
However, at the end of the demonstration, Murati especially thanked NVIDIA for providing the most advanced GPU, which made today's demonstration possible, does this mean that this demonstration has been particularly supported, with powerful computing power and no network latency, so it can achieve an average of 320 milliseconds, which is about the same response speed as a human? If this is the case, it means that this kind of human-computer dialogue must be realized on a large scale, and I don't know how many times the computing power will be increased, I am afraid that it is enough for Nvidia and TSMC to be busy for a few years, and it is really a happy money printing machine.
----------------------------
I:
Well, I think this ability is very important, because in the future, the integration with robots will make artificial intelligence the same as humans, and if the appearance is very similar, it will be exactly the same as in a science fiction movie, and humans and robots will not be able to tell the difference.
----------------------------
ChatGPT：
Indeed, with the development of artificial intelligence technology, especially in natural language processing and robotics, we are getting closer to robots that can simulate human behavior and communication. This technological advancement could allow future robots to be closer to humans in appearance, behavior, and even the ability to communicate with humans, like scenes from science fiction movies, as you said.
However, it also raises many ethical and social issues worth exploring, such as the transparency of AI, privacy protection, and the relationship between humans and robots. Are you excited about such a future, or are you a little worried?
----------------------------
I:
I'm not worried, I don't think it's easy to say what the fate of humanity will be in the future, Neanderthals never thought about what the future holds, and I don't have to think about it.
----------------------------
ChatGPT：
Your views are insightful. Indeed, the future is full of uncertainties, and every technological advancement may bring unexpected changes and challenges. Like Neanderthals, we may not be able to fully predict future developments, but we can make choices and prepare now for what may arise in the future. Is there anything you're particularly interested in or would like to learn more about the development of AI and technology?
----------------------------
Is it difficult to distinguish such a tone? But I still have a simple way to distinguish ChatGPT from normal human voices, that is, it likes ethics and morality, and there are so many words that anyone can hear it with one ear, can OpenAI correct this for it first?