GPT-4 officially launches from OpenAI with text-image support and more

Attachment 63513

As previously rumored, OpenAI has now launched GPT-4, the next-generation version of its chatbot AI program. The blog post that reveals GPT-4 goes into deep detail on the new features and performance improvements.

Attachment 63514

Perhaps the biggest new feature is that GPT-4 can answer inquiries that combine both text and images. The blog states:

Attachment 63515

The bad news is that image inputs are not currently available for the general public as they are in a "research preview" mode. There's no word on when this will be generally available.

The new version will also allow users to alter the tone of GPT-4's answers. The blog states:

Attachment 63516

The blog also showed off some performance charts to illustrate GPT-4's improvements. In one case, OpenAI gave the chatbot some simulated exams. It stated:

Attachment 63517

Attachment 63518

The chart above shows GPT-4 offered much better exam results in most cases compared to GPT-3.5

As with previous versions, GPT-4 is limited in terms of getting more recent information about subjects, as it can only look back to September 2021. OpenAI also says the new version "significantly reduces hallucinations relative to previous models" but it "is not fully reliable."

The GPT-4 API is currently available for developers to access if they sign up for a waitlist. The general public can check out GPT-4 on a limited basis if they sign up for ChatGPT Plus, which costs $20 a month. The blog post mentions OpenAI could launch another subscription tier that will allow for higher volume levels of access.