ChatGPT4, the latest iteration of OpenAI's natural language processing AI model, was released last Tuesday following the success of its predecessor ChatGPT3.5.

OpenAI's ChatGPT4 presents a significant milestone in the field of artificial intelligence, with its advanced language understanding and generation capabilities.

As Malta's technology sector continues to expand, the integration of ChatGPT4 into various industries can lead to innovative solutions, improved efficiency and economic growth.

In a press release issued in tandem with the launch, OpenAI said “while [GPT4 is] less capable than humans in many real-world scenarios, it exhibits human-level performance on various professional and academic benchmarks”.

The company claims that GPT-4 can write code in all popular programming languages as well as read, analyse and synthesise up to 25,000 words of text.

Dr Dylan Seychell at the University of Malta’s Department of Artificial Intelligence, and Mark Erik Szoldan, a member of the Hungarian Engineering Academy and a senior IT specialist, both offered their perspectives on this new and improved version of ChatGPT, and the potential implications this development has for the future.

 

1) What is the difference between the latest version of this AI, ChatGPT4, and the previous iteration ChatGPT3.5?

“Firstly, this new model can handle more words in its user-prompted context than the previous one, meaning that responses can handle a larger volume of text and therefore add relevance to the context that the user adds,” said Seychell.

“Secondly, whereas the previous model received only text input, the new version is multimodal, meaning that it can recognise and process both text and images, although it can only output text for now,” said Seychell.

Seychell clarified that “this ability of AI models to recognise images and detect objects in them has been around since about 10 years ago. What’s new is the merging of this ability to provide more context within the chat feature.”

“Further, the new version also addresses what’s called ‘hallucination’, which is the technical term for when the model invents plausible sounding answers to questions that it doesn’t know the answer to.”

“According to OpenAI, this is an issue that has been drastically reduced,” but Seychell was quick to point out that “this correction isn’t without its dangers too, as people are less likely to fact-check the AI’s claims if they think it’s more likely to be correct in its responses.”

Szoldan was slightly more emphatic in describing the “massive difference between the two versions”, stating that the main difference lies in the latter’s “capability to contextualise the information with which it’s presented”.

Szoldan said that “from the application perspective, I believe this breakthrough is huge because this AI shows an unprecedented ability to understand contexts in a human-like manner. The application possibilities are endless and there are already some ChatGPT4-assisted solutions available. These skills will further improve with time and will develop to be able to understand contexts that we can’t yet fathom”.

“Of course,” he continued, “it will take time for people to start using the technology and for companies to come up with viable services that will be used widely, but, in the longer run, I think it will change a lot of things fundamentally.”

 

2) Does this new version point in the direction of the coming singularity, as proposed by Kurzweil? Or are such notions merely science fiction?

Ray Kurzweil's concept of the singularity refers to a hypothetical point in the future when technological growth, particularly in the fields of artificial intelligence and computing, reaches such an advanced level that it results in rapid, uncontrollable and transformative changes in human society.

Kurzweil, an American inventor, futurist and author, is a prominent advocate of this concept, which he explores in depth in his book The Singularity is Near, published in 2005.

At the heart of the Singularity lies the development of artificial general intelligence (AGI), a level of AI that possesses the ability to perform any intellectual task that a human being can do. 

According to Seychell, “in its current state, ChatGPT4 is definitely not a form of AGI. In fact, it’s still too early to say definitely whether artificial general intelligence is possible or not”.

He added that, despite the fact that we are nowhere near the Singularity with this present technology, the level of refinement in how the information is presented to the end user is exceptional.

Szoldan agreed, stating that “this technology is just interpreting and using the data that’s fed to it, so it’s still not at the stage where it can become fully independent. As such, this technology is not the Singularity”.

“Having said that,” Szoldan said, “I am of the opinion that that point may not be so far away. There already exist some AIs whose sole purpose is the development of further AIs. But if you were to ask me when, and how, and to what extent this point will come, I have to admit that I don’t know.”

 

3) What current uses have you personally found for this new technology in your daily life?

Seychell said that he uses this technology as a writing assistant, “when I’m writing about complex topics, I’ll ask for clarifications on ways of phrasing my points and simplifications for areas that are hard to explain”.

Further, Seychell also uses ChatGPT to tailor lessons in select advanced courses to the needs of every individual student. “Instead of preaching to the students and making them fall asleep, I’m guiding students in their own understanding of the topic at hand.”

In certain lessons, students use ChatGPT to ask specific questions about what they don’t understand. “Essentially, the lesson becomes more of a workshop and I become less of a teacher, as the word teacher is traditionally understood, and more of a facilitator to individual learning strategies,” he said.  

“In those same lessons, we also make sure the students know how to fact-check the information that they’re accessing through the technology,” Seychell said.  

“In my opinion, these are skills that will be obligatory for the society of today and tomorrow. Whoever doesn’t acquire these skills will ‘fall behind’ and, as a university academic, I feel like it’s our responsibility to be among the first to teach the skills required to use this tool adequately,” Seychell said.

“When lecturing the most advanced topics in AI, it’s no longer the case that I know more than the students and they need to play catch-up to my level of knowledge. Ideally, I now act as the guide and together we reach a level of knowledge that supersedes our previous levels of understanding, myself included,” he concluded.

 

4) How do you see this technology changing ‘a lot of things fundamentally’?

Szoldan said that “the AI’s ability to contextualise information will mean that we will be able to simplify certain essential processes that we all rely on and tackle fundamental social problems, such as excessive traffic issues, political corruption and the democratisation of knowledge, to mention a few specific examples”.

Regarding the democratisation of knowledge, “this is a very important point to emphasise because many specialist jobs, in whatever field, are based on the idea that many years of fine-tuning are required for practitioners to function effectively”.

“I believe, that if we’re using this technology right, we will be able to replace this type of specialist.”

“For example, ChatGPT has the ability to understand vast swathes of legal code and intelligently explain to users, not only the law, but also the reasoning behind certain judgments,” Szoldan said. 

To be absolutely sure, Szoldan said that “we would still need specialists to check on the work of the AI’s as they will still make mistakes. But for every kind of service, we would be able to give much quicker, and more accurate, answers to pressing issues, and be able to cut down on excessive and tedious bureaucracy.”

Concerning traffic issues and in addition to existing political measures, “ChatGPT can help ease congestion by massive real-time data analysis to facilitate autonomous driving”.

As for political corruption, Szoldan said that “ChatGPT can reduce the need to rely on the arbitrary whims of political decision-makers. For instance, when it comes to deciding which infrastructure projects to undertake, you can rely on data to make public recommendations based on what needs to be done, which might have a hand in reducing corruption”.

“If properly implemented, essential government services will be so much more efficient. Naturally, we still need to control the AI but with good guidance, it can really have beneficial effects,” said Szoldan.

Szoldan said these results were “not inevitable and required significant political will to implement”.

“With the appearance of various search engines, the challenge was to the question of the reliability of the source of knowledge. Now, with AI-assisted search and given the avalanche of information that we have access to, the question is: how do you phrase your question to the AI to derive the greatest possible benefit from the information?” Szoldan concluded.