ChatGPT and other GPT-based chatbots, like Bing AI, can be useful tools for pretty much anything. However, it’s clear those tools still need a lot of improvement. One huge improvement is coming today, with the launch of GPT-4 by OpenAI.
OpenAI has officially released GPT-4, an improved version of its conversational GPT AI that can, according to the research group, provide more natural, accurate, and creative responses in conversations. It’s 40% more likely to provide factual responses than the previous major version, GPT-3.5, and it’s also 82% less likely to respond to disallowed content. GPT-4 implements changes and feedback from ChatGPT users, which have been using the tool for some time, with the result being a safer-to-use, more accurate AI model.
Perhaps the most impressive improvement is the ability to use images as data input. For example, you could take a picture of the inside of a refrigerator, and ask GPT-4 what recipes can be made using the items. The image recognition technology is similar to what we’ve seen with Google Lens over the years, but fully integrated into the rest of GPT-4’s abilities.
GPT-4 is still not perfect yet, though. Microsoft confirmed that Bing Chat AI has been using GPT-4 all along, and that AI chatbot has gone viral a few times for dumb mistakes and occasionally creepy replies as well. An article from The New York Times showed how GPT-4 tried to describe how words in Spanish are pronounced, which was mostly wrong.
This is probably just a stepping stone in the technology’s journey, and it will keep being improved upon from this point onward. Make sure to check out OpenAI’s official page on GPT-4 for more information — including how to join the API waitlist if you’re a developer.