Finally, Open AI launched the next-generation GPT-4 with various advanced features, including a new multimodal model. As the artificial intelligence company owns popular ChatGPT so, we expected to see this GPT-4 in it first.
But they introduced it in the Bing Chat first, but don’t worry, as it wasn’t filled with all the new features of GPT-4 without many features, so let’s discuss all the details below.
Also Read: How to Access ChatGPT 4 for Free
OpenAI’s GPT-4: All Details
ChatGPT’s success has gained a lot of people and the company’s attention; even Microsoft has fueled its developer company with money to use their technology in their un-popular search engine Bing which has now crossed the 100 million daily active users barrier.
And its developer company decided to work on the next version of its autoregressive language model called GPT-4, and now, it is officially available in ChatGPT, but not for everyone.
As the company keeps it behind a paywall, and if anyone wants to use ChatGPT based on GPT-4 so they have to pay, while the prior ChatGPT based on GPT-3 is still free for everyone.
ChatGPT Plus is the subscription that will allow you to use GPT-4, and it will cost $20 monthly. To upgrade and use ChatGPT, you can go to chat.openai.com, and as developers want to access the GPT-4 API so they can sign up on this waitlist.
Besides, on the same day of its release Yusuf Mehdi, Microsoft’s head of consumer marketing, revealed that their new Bing has already been using GPT-4 for the last five weeks.
And the users who are part of the new Bing preview all have experienced an early version of this powerful model.
Yusuf also congratulated Open AI on GPT-4’s launch and said, “as OpenAI makes updates to GPT-4 and beyond, Bing benefits from those improvements”.
What’s New In GPT-4
According to the developer company, they spent around six months making GPT-4 so that it can generate more acceptable and more uncontroversial compared to GPT-3.5.
It also shared some stats regarding its comparison to GPT-3.5. The new version of GPT is anticipated to not generate results around 82% times for disallowed content, and it is also expected to generate around 40% more factual responses.
It also gained multiple new capabilities, such as generating results in more languages than just English, as they improved around 25 languages.
And GPT-4 is now more capable of clearing exams that are designed for humans with its more trained data and human-like responses.
For example, the company gave it to clear USNCO Local Section Exam 2022, and it scored 36 / 60. However, GPT-3 scored 24 / 60, 10 points less than GPT-4.
Another advancement is visual inputs which we already discovered before its release. With visual inputs, users will be able to give input as text and images. Still, it is going to give output in text form.
You can see the perfect example of this new visual input feature in the above image, where it explains a meme from Reddit.
With all that, it also has similar limitations as GPT-3, which includes “hallucinating” facts on a topic that comes under disallowed content, and it can show reasoning errors. And also, its trained data is still limited to September 2021.
One more best part about GPT-4 is that it includes “Steerability,” which can be explained as different personalities for different situations.
You can understand it by taking a look at the below image in which GPT-4 is being Socratic Tutor rather than just giving the answer.
Now, Microsoft’s Bing is not the only company that is getting benefits from OpenAI’s technology, as it has also partnered with various companies to merge GPT-4 into their products.
And this list includes some popular brand names, such as:
- Duolingo, which is a language-learning company.
- Stripe, which allows electronic payment gateways.
- Khan Academy, which is a popular education company with thousands of courses.
For more information, you can also check out the complete research report from OpenAI that contains various test results of it.