It seems like only yesterday that OpenAI unleashed ChatGPT onto the world and transformed everyone’s view on Artificial Intelligence (AI). In actual fact, it has not been that long ─ less than four months in 30 Nov 2022 ─ since its launch. But now, OpenAI has welcomed a more powerful tool ─ ChatGPT-4.

The war of the AI has only intensified with the latest upgrade while many companies have already jumped on the AI bandwagon. To know more, you can take a look at this article.

This next generation state-of-the-art AI language model represents a significant advancement in AI technology, with the potential to revolutionise various industries. According to The Straits Times, it is "the fourth iteration of the OpenAI software that has analysed vast quantities of information from across the Internet in order to determine how to generate human-sounding text and give users detailed responses to questions.”

You may come across sources calling its predecessor GPT-3 or GPT-3.5. Both are not wrong but to clarify the confusion why some use the former while others use the latter, it is because we are currently interacting with the GPT-3 model and GPT-3.5 is used if you want to get technical.

However, only paid ChatGPT users and developers will have access to it. If you are using the free version, you will need to upgrade by following the steps below. Click on "Upgrade to Plus" then select "Upgrade plan" in the pop-up.

Thereafter, you will be directed to a standard ecommerce sales page to input the necessary details before making the purchase.

Once you have successfully paid, your access to the new model will be immediate. When you initiate a new chat, a pulldown will offer you the option to use one of the old models or the new one. A fun fact: old outputs using GPT-3 will have the old, green OpenAI logo to their left, while GPT-4-powered outputs will have a new, black logo.

On 15 March, Open AI announced the launch via Twitter.

You can replay the developer demo livestream showcasing GPT-4 and its capabilities and limitations here.

The next big question is how does it differ from the previous versions?

ChatGPT-4 builds on the success of the previous versions, including ChatGPT-3, which made headlines for its impressive language capabilities. However, ChatGPT-4 takes this technology to the next level with a larger neural network and more extensive training data.

OpenAI noted that "the new model scores more highly on a range of tests designed to measure intelligence and knowledge in humans and machines…it also makes fewer blunders and can respond to images as well as text.”

The key point here is that this new model can read photos and explain what is in them, in short, a multimodal model (accepting image and text inputs, emitting text outputs), according to a research blog post. The free model we are currently using can only process text inputs. In addition, the ability to input video is on the horizon.

According to OpenAI, "GPT-4 is more creative and collaborative than ever before. It can generate, edit, and iterate with users on creative and technical writing tasks, such as composing songs, writing screenplays, or learning a user's writing style. It can accept images as inputs and generate captions, classifications, and analyses." Capable of handling over 25,000 words, it allows for use cases like long-form content creation, extended conversations, and document search and analysis.

It is definitely an upgrade from the previous version and as OpenAI put it, it is "…safer and more aligned" and "…82% less likely to respond to requests for disallowed content and 40% more likely to produce factual responses than GPT-3.5 on our internal evaluations.”

Not to be a downer or anything but despite its impressive capabilities, ChatGPT-4 still has some flaws and areas for improvement.

On its website, OpenAI pointed out that “GPT-4 still has many known limitations that we are working to address, such as social biases, hallucinations, and adversarial prompts." It also lacks knowledge of events that occurred after about September 2021, when its training data was finalised.

If comparing GPT-4 with previous models, it indeed hallucinates less often, scoring 40% higher than GPT-3.5 in an internal adversarial factuality evaluation. The chart can be seen below.

In terms of accuracy, there may be an improvement but it still makes mistakes. One instance is when an AI researcher and professor, Oren Etzioni, asked the new bot a straightforward question, "What is the relationship between Oren Etzioni and Eli Etzioni?" The response was correct as opposed to the previous version which was always wrong.

However, mistakes were still made when it went on to say, "Oren Etzioni is a computer scientist and the CEO of the Allen Institute for Artificial Intelligence (AI2), while Eli Etozioni is an entrepreneur." It is largely accurate save for the fact that Oren had recently stepped down as the Allen Institute's chief executive.

He expressed that, "While they've made a lot of progress, it's clearly not trustworthy…It's going to be a long time before you want any GPT to run your nuclear power plant.”

In terms of forming hypotheses about the future, it is also less adept; it seemed to draw on what others have said instead of creating new guesses. When Oren asked, “What are the important problems to solve in NLP research over the next decade?" ─ referring to the kind of "natural language processing" research that drives systems' development such as ChatGPT ─ it could not formulate entirely new ideas.

Another issue that haunts all the leading chatbots is the problem with "hallucination". Since the systems do not have an understanding of what is true or false, chances are they may generate text that is completely untrue. As pointed out, when asked for websites’ addresses that described the latest cancer research, it sometimes generated internet addresses that did not exist.

Conclusion
OpenAI's launch of ChatGPT-4 represents a significant step forward in the field of AI language models (GPT3.5 was no less amazing of course). With its advanced capabilities and potential applications, ChatGPT-4 has the potential to change the way we interact with technology. Nonetheless, as with any technology, there are still challenges that need to be addressed. With ongoing monitoring and improvement, ChatGPT-4 can become a powerful tool for improving communication, creativity, and efficiency in numerous industries.

It is already being put to use by clients as diverse as Khan Academy, Stripe, Morgan Stanley, and the government of Iceland. In spite of that, Oren is keen to highlight that as remarkable as GPT-4 is, there are still countless things that we take for granted that it cannot do, "We have to remember that, however eloquent ChatGPT is, it's still just a chatbot.”