ChatGPT-4 is “larger” than earlier iterations, i.e. it has been trained on more data and has more weights in its model file, which also means it costs more to run.
Many artificial intelligence researchers today believe that recent developments result from running increasingly large models on thousands of supercomputers in training processes that can cost tens of millions of dollars. GPT-4 is an example of an approach focused on “scaling up” in order to produce better results.
OpenAI said it used Microsoft Azure to train the model (Microsoft has invested billions in the startup). The company cited “the competitive landscape” in the market as justification for withholding information regarding the precise model size or the hardware it trained with.
OpenAI’s GPT language model is behind many of the AI demos that have been astounding people in the tech world over the past six months, including Bing’s AI chat and ChatGPT, and the latest version previews new improvements that could gradually have an impact on consumer products like chatbots in the weeks to come. For instance, Microsoft confirmed that Bing’s AI chatbot already uses GPT-4.
GPT-4 can solve difficult problems with greater accuracy, thanks to its broader general knowledge and advanced reasoning capabilities.OpenAI
The creator of the language model guarantees that the new generation will produce fewer factually incorrect answers, veer less often into forbidden topics, and even outperform humans on numerous standardized tests. They claim that the GPT-4 scored in the 90th percentile on a simulated bar exam, the 93rd percentile on SAT reading exam, and the 89th percentile on the SAT Math exam.
However, OpenAI cautions that the new software is far from perfect and is still less capable than humans in many contexts. It still has a huge problem with making things up and not being completely factually reliable:
GPT-4 still has many known limitations that we are working to address, such as social biases, hallucinations, and adversarial prompts.
“In a casual conversation, the distinction between GPT-3.5 and GPT-4 can be subtle. The difference comes out when the complexity of the task reaches a sufficient threshold – GPT-4 is more reliable, creative, and able to handle much more nuanced instructions than GPT-3.5,” OpenAI wrote in a blog post.
The new model will be made available to paid ChatGPT subscribers and as part of an application programming interface (API) that enables developers to incorporate artificial intelligence into their apps. For every 750 words of prompts, OpenAI will charge roughly 3 cents, and 6 cents for every 750 words of response.