OpenAI Spring Update – More Natural Chatbot
For about a week, it looked like OpenAI was ready to take on Google and announce a ChatGPT-based search engine . However, in this case, the rumor mill got it all wrong. Instead, during the Spring Update event earlier today, OpenAI introduced some modest updates to the underlying ChatGPT model, but in a surprising and sometimes disturbing way.
Introducing GPT-4o, the new flagship model
The big announcement from OpenAI was the new GPT-4o model. In turn, the company said that GPT-4o is not only for paying customers – it is available to everyone for free. The company sees GPT-4o as the first step toward making interactions with AI more “natural,” a position that made sense during the presentation.
GPT-4o works with voice, text and vision, so you can interact with ChatGPT using any type of content you like. In addition, OpenAI makes many of its premium features free for everyone. Free users can access GPT for the first time through the GPT Store, upload and chat about images (documents or images) with ChatGPT, and access ChatGPT’s memory feature. The latter is especially useful: ChatGPT will remember what you talked about in past chats, so your future chats will be based on those conversations.
Paid users still have a five-time capacity limit, so there’s something there that makes spending that $20 a month worth it.
OpenAI demonstrated the new model by demonstrating a breathing exercise. The demonstrator asked ChatGPT for some relaxation tips, including instructions to breathe deeply. The demonstrator then began to breathe rapidly and loudly, trying to see if the model would identify the incorrect technique. Indeed, the model corrected the behavior, but it was a little intermittent: the model kept interrupting and interrupting, giving feedback on breathing technique. However, you can “naturally” interrupt a model’s speech, so it’s possible that the demonstrator accidentally interrupted her.
From there, demonstrators asked ChatGPT to come up with a story. It started off as you’d expect from ChatGPT, but one of the demonstrators interrupted him, asking him to add more emotion to his voice. Honestly, it was impressive how the voice model began to act like a voiceover artist from a cartoon, especially when she was asked to emote for the second time. When asked, he even started talking like a stereotypical robot.
I was a little confused when the demonstrators showed how you can feed ChatGPT a live camera feed to analyze your surroundings. They used a simple math homework example, but I don’t know if I’m ready for ChatGPT to have constant access to my environment. If I want to ask him a question about something in front of me, an image or video will do. To prove my point, during this part of the demo they tried to turn off the model, but it suddenly said something like “wow, that’s exactly what you’re wearing.” Yeah, I’m really not here for the AI livestream.
It can also recognize facial expressions on live TV, which again: creepy. One demonstrator showed their face in the feed and asked what they looked like, and ChatGPT said something like “piece of wood,” which the demonstrator quickly corrected by saying it was in response to an image he had previously sent to the chatbot. . (January, of course.) When he tried ChatGPT again, he was able to identify his facial expression.
GPT-4o can also perform live translations, which the team demonstrated live. One person pretended to speak only Italian, while another said he only spoke English: As far as I could tell, the live translation worked well: ChatGPT spoke Italian, and I have to take OpenAI’s word for it that everything that he said was correct.
Based on the demo, GPT-4o will be launching in the next few weeks and I’m looking forward to testing it out. Until then, I’m a little nervous about this experience. The voice effects are quite realistic, and at times everything seems quite natural, but completely unnatural. ChatGPT will experience “human” moments, such as when you say something “oh, stupid me” or “well, that makes more sense” after a correction. Sure, it’s impressive, but I’m not sure I want this technology in my life. What’s wrong with computers being computers themselves? Why would I pretend that my AI is actually alive? Anyway, I’m not keeping this live stream public.
There’s also a new ChatGPT desktop app.
While GPT-4o has been in the shadows, OpenAI also announced a desktop app for ChatGPT, as well as a new user interface, but didn’t go into too much detail about the changes.
The app is similar to the web and mobile version of ChatGPT, but has some new features. Demonstrators showed off the voice application built into this version of ChatGPT; he can’t see anything on your screen, but you can talk to him in the same conversational way. In the demo, they copied the code into the voice app and ChatGPT analyzed the code and explained it as you would expect.