The next generation of ChatGPT, a groundbreaking leap in AI technology, is here. But is it worth the financial investment? Let’s review some key differences between ChatGPT3.5 and ChatGPT4 and delve into what you need to understand.

ChatGPT, OpenAI’s powerful Natural Language Generation (NLG) tool that autonomously generates text, rocked the tech scene in late 2022, much akin to the impression left by its sibling, the Dall-E image-creation AI earlier that same year. The chatbot-like Large Language Model has relied on tons of data for training, from CommonCrawl, Wikipedia, sets the company calls “Books” and some web crawling. Now, OpenAI’s text generation technology has evolved into its 4th iteration, known as GPT-4, an acronym for Generative Pre-trained Transformer that might even put an Autobot to shame. But is this new technology accessible for use yet? And if so, why would you want to utilize it? Here are the answers.

So, What’s the Buzz Around GPT-4?

OpenAI has been steadily releasing iterations of GPT for nearly half a decade. The inaugural public release happened in 2020, igniting a flurry of AI-related announcements from industry heavyweights, including Microsoft, which later became an investor in OpenAI.

As TechTarget explains, parameters are “components of a large language model that determine its proficiency in tasks such as text generation.” They essentially encapsulate what the model learns. GPT-1 was designed with 117 million parameters, GPT-2 boasted 1.5 billion, and GPT-3, which debuted in February 2021, had a staggering 175 billion parameters. By the time ChatGPT was made publicly available in November 2022, the technology had evolved to version 3.5. However, if you’re using the complimentary version of ChatGPT, you’ll be interacting with the GPT-3.5 model for a little while longer.

How Can I Try GPT-4?

ChatGPT quickly garnered popularity, causing server capacity issues. To manage this, OpenAI, the organization behind the technology, introduced a paid version, which didn’t do much to slow its traction. At its peak, ChatGPT (both the free and paid versions) generated web traffic comparable to that of Bing’s search engine. Even now, there are instances when the basic ChatGPT capacity is exceeded – a situation I encountered firsthand while composing this article.

The premium version is known as ChatGPT Plus (or ChatGPT+), priced at $20 per month. OpenAI rolled out a pilot for ChatGPT Plus in early February, which became globally available on February 10. ChatGPT+ is now the primary channel for users to experience the advanced GPT-4 technology.

To get started, you’ll require a free OpenAI account, which you may already possess if you’ve used Dall-E to create AI images. If not, you’ll need to sign up. Then, click on the ‘Upgrade to Plus’ option in the menu. However, capacity issues may occasionally prevent sign-in, a problem ChatGPT+ aims to alleviate.

It’s worth noting that an OpenAI account isn’t the only avenue to access GPT-4 technology. GPT-4 also powers Quora’s Poe Subscriptions, and the company is collaborating with Claude, Anthropic’s “friendly, honest, and harmless” AI chatbot competition.

Microsoft’s Bing search engine, an early adopter of OpenAI’s technology, has been running a customized version of GPT-4 for search. To access it, sign up for the New Bing preview using the Microsoft Edge browser. There’s no longer a waitlist, and any updates to ChatGPT-4 will be integrated into the search engine. Interestingly, when I inquired whether Bing uses GPT-4, it cheekily responded, “No… I’m using Bing’s own natural language generation system. 😊” One advantage of the Bing version is its unrestricted dataset: GPT-4 and ChatGPT+ only use data up to September 2021.

Other organizations and services utilizing GPT-4 include the Icelandic government, Duolingo, and Khan Academy.

Moreover, OpenAI has made the application programming interface (API) for GPT-4 accessible to developers, so anticipate its integration into other services soon.

So, is ChatGPT+ Worth the Hype and the Money?

Based on anecdotal evidence and OpenAI’s impressive statistics, it appears that anyone who has interacted with ChatGPT on the GPT-3.5 version will find GPT-4 exceptional. For instance, GPT-4 has outperformed its predecessor in various exams such as LSAT, SAT, Uniform Bar Exam, and GRE. OpenAI states that, in comparison to GPT-3.5, GPT-4 is 82% less likely to respond to technically impermissible prompts and 60% less prone to inventing false information, a phenomenon known in AI as “hallucinations.” Notably, in a test run by the non-profit Alignment Research Center, GPT-4 was even able to convince a real human on TaskRabbit to perform a task, effectively bypassing a CAPTCHA.

A key distinction between the two versions is that while GPT-3.5 is a text-to-text model, GPT-4 has evolved into more of a data-to-text model, capable of tasks beyond the imagination of the prior version. This infographic details some of the other differences.

For example, GPT-4 can take images as part of a prompt. In one instance, it was shown an image of the inside of a refrigerator and proceeded to generate recipes based on the ingredients identified. It even has the capacity to elucidate why memes are amusing, qualifying GPT-4 as a “multimodal model.” However, for the time being, ChatGPT+ will continue to only output text.