Hengyu from Aofei Temple

Quantum Bit | Public Account QbitAI

Futures fulfilled!!

just,GPT-4o’s voice capabilities, the much-anticipated Her, has finally begun to be opened to users!

OpenAI official said that the selected Alpha users willChatGPT An email with instructions was received in the app.

Like this：

Selected Plus Friends(That is, boughtOpenAIMembers), you can experience interruption at any timeGPT-4oSpeak, or have it sense and respond to conversational emotions.

I randomly picked a netizen’s actual test and posted it here. It’s so funny that I can’t stop laughing!

Without further ado, I sincerely invite you all to enjoy:

OpenAI also stated that the backend will gradually expand the scope of Alpha testing. At what speed?

Around fall:

In principle, the plan is to allow every Plus user to use advanced voice functions that can listen, speak and see this fall.
Video and screen sharing capabilities will be available later.

Currently, Her only provides 4 preset voices to speak, and has built a system to block plug-ins from adding other voices.

At the same time, protective measures are implemented to block output requests for violent or copyrighted content.

It makes our netizens anxious.I can't wait to become the chosen one：

I don't want to listen to this bullshit. I don't want 40 security, I want it to be usable right now!

Another netizen revealed his domineering side when he told OpenAI that "Alpha testing will make us strive for excellence":

Let me participate in the Alpha test, you can learn more from me.

So--

Has anyone started playing?

Of course! Some people have already started showing off their GPT-4o advanced voice capabilities on YouTube or Twitter.(Hereinafter referred to as Her)The happy playing process.

To netizens(And qubits)I'm so envious.

Video function, let’s go

Most of the netizens who have received Alpha qualifications are still in the primary exploration stage, and this timeVideo mode is available only to individual users.

The video content is roughly that Brother M just got a kitten, and he asked Her to help him look at everything about the kitten's life:

After experiencing it firsthand, the guy exclaimed: "It feels like I'm communicating face to face with a knowledgeable friend!"

Serve as a foreign language teacher

For example, this guy named A B on YouTube, who was eager to learn, immediately invited 4o to practice French with him.

He first asked Her whether he pronounced "croissant" correctly.

As expected, Her adopted the method of encouraging education.(Of course, Brother B himself is a good reader), even boasting "very close!"

After watching this video, netizens were so greedy that their saliva started to flow from the corners of their eyes.

But as for Her’s spoken English, this French native speaker gave her full approval, “I can attest that all the pronunciation suggestions are great!”

Listen to jokes and be the supporting role

At the request of netizens, netizen KK told Her a joke.

On the one hand, it is to test whether it can understand jokes, and on the other hand, I want to hear whether its laughter is natural.

The following is a transcript of the entire joke:

Username KK：Can I tell you a joke?

Her：Of course! I really want to hear it. What's your joke?

Username KK：Why don't scientists believe Adam?

Her：why?

Username KK：Because he constitutes everything (here is a pun,Adam—Atom）。

Her：Great, hahahahahahahahahahahahahaha, Adam/atoms really do make up everything!

After listening to the joke, netizens gave enthusiastic comments, saying they really liked the sound of the AI ​​after hearing the joke!

A ventriloquist who can tell stories

Most of my friends can’t play it themselves, and can only rely on netizens who have obtained Alpha qualifications to play it.

So, someone asked KK: Can Her tell a story? The kind with wind, rain, thunder, and atmosphere?

Brother KK responded to the request and quickly posted the test video on Twitter.

Her wrote a story about an AI robot encountering a lady on a rainy night. The story content was in line with the "science fiction elements" and there was nothing wrong with it.

What’s interesting is that in order to meet the “sense of atmosphere” requirement, Her came up with a lot of onomatopoeia to describe the sound of thunder, boots stepping on puddles, etc.

This part is really hard to describe in words, please listen to it and feel it for yourself:

Earlier presentation at CVPR

Finally, I would like to share here a hands-on demonstration of GPT-4o by OpenAI multimodal researchers at CVPR more than a month ago.

The brother asked：Guess where I am?
Her answer：After reading it, I feel like you are at CVPR~

There was a burst of "wow" from the audience.

One More Thing

After a tour, everyone was having a great time. In addition to the gameplay we shared with you above, some people asked Her to pretend to be a football commentator or dub "Romeo and Juliet" or something like that.

Although many people have not yet developed its "sightly" side, they still have a lot of fun playing with it.

Everyone is so happy, can you take me with you???

Also, in addition to issuing GPT-4o advanced speech quotas, has OpenAI, which has always been criticized for not being open, announced anything else?

It only used one sentence to lay the groundwork:

"usPlans to share a detailed report on GPT-4o’s capabilities, limitations, and safety assessments in early August。”

Fine, Fine, talking about functions and limitations, etc., seems unlikely to be the technical report that everyone is thinking of.

But what else can we do? Just wait and make do.

What else can we do?

Reference Links:
[1]https://twitter.com/tsarnick/status/1818402307115241608
[2]https://x.com/yoimnotkesku/status/1818406786077970663
[3]https://x.com/yoimnotkesku/status/1818415019349901354
[4]https://x.com/manuvision/status/1818412120373182928?s=46&t=iTysI4vQLQqCNJjSmBODPw