GPT-4 Is here to change the AI landscape! Here’s how it compares against ChatGPT.
OpenAI, the creator of ChatGPT, has launched the next version of the underlying technology behind ChatGPT, GPT-4. It comes with flashy new features such as image inputs and steerability. It is also said to be more accurate and reliable.
You might be wondering what it is and how it differs from ChatGPT. Let us discuss in detail what GPT-4 is, how it is connected to ChatGPT and the differences between them in this article.
What is GPT-4?
GPT-4, or Generative Pre-trained Transformer 4, is a large language model created by OpenAI. It is the next generation of GPT, which advances the underlying technology used by ChatGPT, which is based on GPT-3.5.
GPT-4 launch was announced by Microsoft last week, to everyone’s surprise. Everyone was already stunned at what GPT-3.5-based ChatGPT could do; it is actually scary what GPT-4 based ChatGPT could do. We will talk about the features of GPT-4 soon, but before that, let’s first discuss how it is different from ChatGPT.
How Is GPT-4 Different From ChatGPT?
OpenAI released the first version of GPT in 2018 and has released various versions over the years. GPT-3 was released in 2020, and the improved version of it, GPT-3.5, was released in March of last year. That leads us to the ChatGPT launch in November last year, which used GPT-3.5.
ChatGPT is a conversational chatbot that uses its underlying large language model of GPT-3.5 to provide complex answers to your queries.
Now, GPT-4 is the latest version of GPT, and soon, ChatGPT will be based on GPT-4. ChatGPT based on GPT-4 is already available for ChatGPT Plus subscribers.
We have seen that we can’t really compare GPT-4 against ChatGPT, as GPT is the underlying technology powering ChatGPT. And GPT-4 will be the technology ChatGPT will be based on. To find the difference between the existing ChatGPT and GPT-4, we have to compare it against GPT-3.5, its previous version.
GPT-3 vs GPT-4: How Is The New Version Better?
On the announcement of GPT-4, OpenAI revealed what GPT-4 is capable of doing. Let’s look at that and compare it against GPT-3.
As you might have guessed, GPT-4 performs better than its predecessor.
It can now handle more data; GPT-3.5 handled 4,096 tokers or around 8,000 words of data. GPT-4 has a big jump in those numbers; it can handle 32,768 tokens or 64,000 words. This means that GPT-4 can generate much longer responses than GPT-3.5, allowing for more detailed responses.
GPT-4 Can Take Image or Visual Inputs
The new GPT-4 can take both text and images as inputs. It can analyse the image provided and the prompt to provide an output text. GPT-4 can handle any type of image, be it an image of a math problem (handwritten even); it can even describe everything shown in a screenshot and even write code for a hand-drawn mockup of a website you want to create.
Simply said, GPT-4 is both a language and visual model, while GPT-3.5 is strictly a language model.
Currently, the image part isn’t available on ChatGPT Plus, but it will be coming soon to the service and possibly to free users as well.
OpenAI has provided a few examples for demonstrating this. Let’s take a look:
Example 1: Asking GPT-4 what’s funny in an image.
Example 2: Deducing the sum of average daily Meat consumption based on a chart.
Example 3: Answer the question given on a question paper.
OpenAI says it has been working on defining the behaviour of AIs.If you’ve used ChatGPT, you might be familiar with its outputs and how it mostly replies with a similar tone. It is because the tone, style and verbosity are fixed, and it stays true to the rules for the most part. With GPT-4, OpenAI is aiming to improve on that. Developers and soon ChatGPT users can set the tone and style of the outputs from the AI.
Take a look at these examples:
Example 1: Acting as a socratic tutor.
Example 2: Acting as a Shakespearean pirate.
Example 3: Acting as a JSON AI assistant.
GPT-4 Is More Factually Correct Than GPT-3.5
While ChatGPT based on GPT-3.5 gives replies for any and every query you give, it tends to hallucinate. In AI terms, this means the tendency of the AI to produce non-sensical and untrue statements presented as facts.
Hallucination is still a problem with GPT-4, but OpenAI says that GPT-4 is 19% to 29% less likely to hallucinate than the older GPT-3.5 model.
GPT-4 Is More Multi-Lingual
While ChatGPT is predominantly used by English speakers, it can also handle other languages. But with GPT-4, OpenAI demonstrated that it could take questions with high accuracy in 26 languages.
The new model also has improved fluency in various languages, making it more accessible and useful to users worldwide.
How To Get Access To GPT-4?
By now, you might have gotten the idea that GPT-4 is superior to GPT-3.5 and are wondering how to access GPT-4; let’s talk about that.
GPT-4 is currently available on ChatGPT on its paid version, ChatGPT Plus. It costs $20 per month, and it is how you get access to GPT-4 LLM primarily. GPT-4 will be available soon to all users of ChatGPT; if you can’t wait, you need to pay for the service to get access.
However, ChatGPT Plus is one of many ways to access GPT-4. Quora’s Poe subscriptions also offer GPT-4 powered chatbots. It also provides another AI model named Claude+ from Anthropic with the same subscription.
But there’s a free way to access GPT-4, and it has been in front of our eyes the whole time. Just after the announcement of GPT-4, Microsoft announced that its New Bing chat was powered by GPT-4 all along. Microsoft previously mentioned using a much more powerful AI, codenamed Prometheus, which was much more powerful than GPT-3.5. Now we know that was just an unreleased GPT-4. You can now use Microsoft’s Bing chat without joining any waitlist, Microsoft has recently removed the need for a waitlist, and you can access GPT-4 on Bing after signing into your account.
Abhijith has been writing about Tech since 2013 on his own blogs. He handles Tech news in KnowYourMobile. He also used to host a YouTube channel about phones. Other than Tech, he is into Formula 1, Wrestling, various TV shows and buying Kindle books he barely reads.