The next-generation big language model known as GPT-4, which serves as the technical underpinning for both ChatGPT and Microsoft’s Bing AI chatbots, was unveiled by OpenAI today. It’s a significant improvement that might pave the way for important improvements to the features and capabilities of current AI technology.
This morning, OpenAI posted on its blog about the new GPT-4 upgrade, which is already available for testing on ChatGPT, the AI chatbot that OpenAI has made available online. (But only the premium version, ChatGPT Plus. Moreover, there is a cap of 100 messages per four hours.) Microsoft also acknowledged that the GPT-4 protocol serves as the foundation for the most recent version of their Bing Talk feature.
Why is this important? The largest improvement that GPT-4 brings is the first on our list of significant enhancements: It’s no longer simply about words.
GPT-4 may now view the outside world through visual input.
According to OpenAI, GPT-4 can take photos as inputs and produce captions, classifications, and analysis. Well, what you think it means is true: ChatGPT and Bing will be able to “see” the world around them or, at the very least, interpret visual results in a similar manner to how image search presently works.
What exactly does this mean in practice? It basically implies that GPT-4 will truly view the real environment, just like it does for apps like Be My Eyes, a tool for people who have vision problems. Be My Eyes uses a smartphone’s camera to illustrate what it is viewing.
Greg Brockman, president and co-founder of OpenAI, demonstrated how GPT-4 translated a design, changed it into a Web site, then provided the code for that site in a GPT-4 video geared at developers.
The extended output of GPT-4: Epic fanfiction
According to OpenAI, “GPT-4 can handle over 25,000 words of text, enabling use cases including long-form content creation, lengthy dialogues, and document search and analysis. Pray for the moderators who oversee Kindle self-publishing, as my colleague Michael Crider phrased it.
Why? Since ChatGPT’s output with GPT-4 will expand in both length and creativity: More collaboratively and creatively than ever before, according to OpenAI, is GT-4. It can generate, edit, and collaborate with users on technical and artistic writing tasks, such authoring screenplays, music, or learning a user’s writing style.
But in reality, there’s another advantage. It appears that you will be able to feed additional web pages directly into the prompt based on the examples that OpenAI provided. Certainly, GPT-4 ought to be familiar with the internet, yet GPT-3 had training till 2021. If it isn’t already current, it appears that you will be able to provide GPT-4 and ChatGPT with greater context.
GPT-4 is merely wiser.
Big language models are not intelligent by nature. Yet, they do comprehend the connections between words, and the more complex GPT-4 model will comprehend connections and context much better. As an illustration, ChatGPT passed the universal bar test in the 10th percentile. GPT-4 achieved a 90th percentile score. The vision-powered GPT-4 achieved a score in the 99th percentile in the so-called Biology Olympiad, while ChatGPT came in at the 31st percentile.
According to OpenAI, Duolingo also uses GPT-4, which enhances its contextual knowledge of what you’re saying and how you need to be saying it.
Including doing your taxes with it? I guess, sort of. Brockman entered the tax code, and the GPT-4 software generated a fictitious tax liability.
