Update Check this out! ChatGPT just got a major upgrade, man. Now you can upload images, talk to the chatbot, and actually hear it talk back, dude.
OpenAI dropped the news on Monday that these awesome new features will be available to paid subscribers and enterprise customers in the next two weeks on their web, iOS, and Android apps. And don’t worry, my friend, they’re gonna roll it out for the free version later too, man.
But hold up, with these new capabilities also come some concerns. OpenAI wants to make sure people don’t start using it for all the wrong reasons, like generating inappropriate or offensive remarks. So, they’ve put some restrictions on commenting about certain types of images, man.
Now listen, these vision-based models, man, they can be tricky. Sometimes they start seeing things that ain’t even there, like people who aren’t even present. And in certain important areas like extremism and scientific proficiency, OpenAI wanted to make sure this technology is ready for prime time, you know? So they tested it with some experts and got their input, and that’s how they made sure it’s responsible and all that, dude.
And listen, man, OpenAI knows that people’s privacy is important. So they’ve taken technical measures to make sure ChatGPT doesn’t go around analyzing and making personal statements about people. ‘Cause, you know, it’s not always accurate and respecting privacy is crucial, bro.
But here’s the really cool part, man. Now ChatGPT can do so much more with different types of data, not just text. You can upload images of cool things like historical landmarks and let the chatbot teach you all about them. Or, check this out, you can show it pictures of the inside of your fridge and tell it to whip up something delicious with the ingredients you have. And get this, you can even highlight specific parts of an image for the chatbot to focus on, dude.
Oh, and here’s another mind-blowing update, bro. OpenAI hooked up ChatGPT with a speech recognition model called Whisper, man. So now you can talk to the chatbot and it’ll transcribe your voice into text. And that’s not all, they even added a new system for the chatbot to speak back to you, man. And get this, you can choose from five different AI-generated voices, dude.
And you won’t believe this, man. Spotify is even using this cutting-edge audio model to translate podcasts into different languages, all while keeping the original voices intact. That’s some serious next-level stuff, bro.
But hey, hold up. Right now, ChatGPT can only transcribe speech in English, man. So if you start talking to it in a different language, especially one that doesn’t use the Latin-based alphabet, it might struggle a bit, you know?
Now, let me drop some knowledge on you, man. These large language models, they’re powerful, no doubt. But they ain’t perfect, dude. Sometimes they come up with false information, and you definitely don’t wanna rely on the chatbot for risky decisions, like figuring out which mushrooms are safe to eat, for example. Remember what Sir Terry Pratchett said – “All Fungi are edible. Some fungi are only edible once.”
The Register reached out to OpenAI for some more details, man. Like, will OpenAI be collecting users’ voices and images? That’s an important question, bro. ‘Cause OpenAI previously mentioned they won’t be using data from enterprise customers or people’s conversations if they disable chat histories. So we’ll see what they say, my friend. Stay tuned.
Updated to add
Hold up, man. OpenAI just confirmed something. They’re gonna use data from “non-API consumer services ChatGPT or DALL-E” to train their models, unless the user chooses not to, bro. Same goes for Whisper, man.