Chat GPT can now speak and sing in real time | DW News

The AI race has just shifted into high gear, with US artificial intelligence pioneers OpenAI rolling out its new interface that works with audio and vision as well as text. The new model, called GPT-4o, has gone beyond the familiar chat-bot features and is capable of real-time, near-natural voice conversations. The developer OpenAI will also make it available to free users.

ChatGPT was already able to talk to users, but with long pauses to process the data. It often seemed a bit sluggish. This was because the feature required three internal applications, the company explained: transcribing the spoken text, processing and generating, and converting the response to speech. This caused delays.

We talk to computer scientist Mike Cook from the renowned Kings College London about the new Chat GPT-4o development.

#artificialintelligence #chatgpt #openai

Follow DW on social media:

[omegaRST]

This is one of the best interviews that I have seen on this topic, great job DW

[krlscape]

this is such a good feature for people with low vision

[Jeevanm]

RIP tour guides, translators, tutors etc

[jrbb]

We've come a long way from hotdog and not hotdog

[williamlai]

Random guy: That girl is pretty, should I date her?
ChatGPT: She's above your paygrade.
Random guy:

[selam]

Instead of fear mongering let’s stop and ponder and celebrate what we just witnessed in the video with the blind guy. 🎉

[pavlinpetkov]

A lot of people will get axed... This kind of rapid progress is unknown in human history. People do not have time to adjust to the changes.

[Loveyogaanatomy]

such a nice intelligent and clear speaker on the subject

[PaulieRubinDMize-uulc]

Seriously sounds like Scarlet Johansson

[Nainara]

ChatGPT and Microsoft's copilot have probably made my team about twice as efficient. Moreover, it's really expanded the "comfort zone" of my colleagues in terms of the computer languages and technology domains that they're mentally prepared to grapple with.

[grahamashe]

Wow. That was a really good interview.

[headofmyself]

Good analysis that explains why they made it free. The model works natively also now with Audio and Images. That means imho that they can tokenize this data directly and then feed it into the transformer architecture. Now, whilst the current versions understanding of the world was based on free internet data, they can now use much, much more data of the real world in order to train the models, resulting in really powerful future models. And of course, it is your data you feed into to this. Thats the scary part.

[bthrkay]

Fantastic interview guys 👏 smart questions and very well spoken answers

[MoonlightShadow]

In the end, everything can be quantified using statistics as long you know how to fitting the right function.

[kbboy]

You can feel she's concerned she might loose her job.

[ryanmckenna]

I cannot believe how fast this is moving forward.

[KkeoPPhachith]

Tour guides are not required anymore when you’re at a museum.

[aliettienne]

7:28 This is just the beginning of what will be highly transformative to our modern world. The fact that audio, image and video can merge together into one to give us human-like interaction is just phenomenal. The pace of Ai progression is beginning to accelerate. 😎💯💪🏿👍🏿

[futurestyle]

For your information, the world's first image based on the XFutuRestyle algorithm using GPT-4 was created in Ukraine and presented at the international exhibition of digital art in London and Athens, which drew OpenAI's attention to Ukraine's technological potential

[Kneephry]

The commentator mentions the risk of rapid adoption of this technology in education or healthcare but it's worth noting there's risk in slow adoption as well. It could be this technology saves lives in healthcare or improves education. I'm not saying we throw caution to the wind but we also shouldn't be so cautious that we slow beneficial technology too long.