Covering Disruptive Technology Powering Business in The Digital Age

By: Martin Dale D. Bolima

OpenAI’s highly successful ChatGPT-3 now has a big brother: ChatGPT-4.

ChatGPT-4 is the latest iteration of the world’s most advanced large language model (LLM) and is being touted as a significant enhancement of GPT3.5, the LLM powering the now very popular ChatGPT-3. According to various reports, ChatGPT4 will boast a range of new features, the most notable being a multimodal capabilities. This feature will purportedly enable the newest chatbot AI to work on text, image and even video inputs.

According to the OpenAI website, ChatGPT-4 “is more creative and collaborative than ever before” and “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.”

Reports also indicate that ChatGPT-4 has been trained on a massive dataset of over 100 trillion words—as much as 10 times larger than the dataset used to train ChatGPT-3. This more comprehensive “learning” means ChatGPT-4 will, in theory, have an even deeper and more nuanced understanding of language, thus allowing it to generate deeper, more sophisticated responses.

ChatGPT-4 is also said to be much more comprehensive than its younger sibling, ChatGPT-3. According to OpenAI, the creators of both ChatGPT-3 and ChatGPT-4, the newest iteration of its AI chatbot can respond using up to 25,000 words, or as much as 700% more than the current 3,000-word limit for the free version of ChatGPT-3. This increased capacity means ChatGPT-4 will be able to provide more context and greater nuance in its responses. It also means it will be able to handle larger text inputs, summarise entire articles or websites and even longer text forms.

Equally important, ChatGPT-4 is said to be safer and more accurate than all its GPT brethren.

“We spent six months making GPT-4 safer and more aligned,” wrote the OpenAI team on their official website. “GPT-4 is 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.”

The OpenAI are also touting ChatGPT-4’s “improved behaviour, which they claim is the result of incorporating “more human feedback, including feedback submitted by ChatGPT users, to improve GPT-4’s behaviour” and working with “over 50 experts for early feedback in domains, including AI safety and security.”

ChatGPT-4, like its predecessor, will also be learning on-the-fly, with OpenAI promising to update and improve GPT-4 “at a regular cadence as more people use it.” Meaning, it can technically “get better” as more and more people provide their inputs and feedback.

But while ChatGPT-4 is a massive leap forward for GPT, OpenAI itself acknowledged the limitations of its newest LLM. ChatGPT-4, like previous GPT iterations, can still be weighed down by “social biases, hallucinations and adversarial prompts,” leading to some prejudiced, contrarian or head-scratching responses.

In other words, ChatPT-4 is still not perfect. It is far from it, in fact. But it builds upon the successes of its predecessor, ChatGPT-3, which was released in 2020. Since then, ChatGPT-3 has become the gold standard of AI chatbots with its ability to generate highly realistic language complete with grammar and syntax. With ChatGPT-4, a new gold standard is likely to emerge.

(0)(0)