Artificial intelligence text generation programs broke into the public consciousness late last year when OpenAI released a free prototype of their artificial intelligence chatbot called ChatGPT.
OpenAI has now released its newest version of their artificial intelligence model, called GPT-4.
People said GPT-3 would be the death of the writer, but . . . it wasn’t.
Although it’s an incredible technological leap, GPT-3 is full of flaws. It gets facts wrong, its writing is often dry and clinical, and sometimes it strays off-topic into irrelevant ramblings.
It also lacks the ability to write about topics that aren’t in the public domain, like a random dude’s biography, for example.
There’s no doubt that GPT-4 will be better but will it be the death of the writer this time around?
What is OpenAI and GPT?
OpenAI is an artificial intelligence company founded by Sam Altman and other Silicon Valley investors, including Twitter C.E.O. Elon Musk, in 2015.
Most of OpenAI’s work went on outside of the public eye, but it was clear to those in the industry that OpenAI’s technology was extremely promising.
In 2018, OpenAI became so large that Elon Musk stepped away from the company over fears of conflicts of interest with Tesla who was developing their own artificial intelligence programs.
Microsoft then invested $1 billion into OpenAI in 2019. In 2020, OpenAI entered into an exclusive licensing agreement with Microsoft.
OpenAI first entered the mainstream in 2021 with an artificial intelligence-powered image generation program called DALL-E.
A free version of this program, named DALL-E Mini (now Craiyon), was made available to the public. OpenAI used a similar strategy when they revealed their text generation program.
OpenAI released ChatGPT to the public in November 2022 while also offering a paid version to other tech companies and individuals.
ChatGPT allows users to enter in a prompt and receive human-like text generated through a complicated artificial intelligence model.
The model operates like a chatbot, allowing for users to communicate with the model to suggest edits that the model would then correct.
Users flocked to OpenAI’s website to test out the revolutionary technology. Users were required to create a free account with OpenAI to access the ChatGPT tool.
It took just one week for ChatGPT to earn one million subscribers. For context, it took Twitter and Netflix years to cross the one million subscribers threshold.
As screenshots of ChatGPT-generated text flooded social media feeds, many were left stunned by just how well the program was able to produce text nearly indistinguishable from text written by a human.
To make this human-like text, ChatGPT uses OpenAI’s Generative Pre-Trained Transformer 3 (GPT-3). GPT-3 is the successor to GPT-2.
It represents a major jump in artificial intelligence text generation capabilities. GPT-3 was first released as a beta in July of 2020.
GPT-3 relies on a deep learning artificial intelligence model which is able to process information produced by humans and replicate it when fed a prompt by humans.
Its model is fine-tuned by the humans at OpenAI to reinforce good answers, e.g., the most human-like answers. Programmers also discouraged “bad behavior,” such as racial bias, which often plague artificial intelligence programs.
The result of this fine tuning, as well as an expanded and updated database for the artificial intelligence to draw from. OpenAI has released the brand new GPT-4.
What is GPT-4 and how is it different from GPT-3?
GPT-4 is an updated version of OpenAI’s GPT-3 text generation model. GPT-4 is larger, faster, and even more human-like than GPT-3.
Artificial intelligence language models rely on learning from databases of information fed into them. Artificial intelligence models are also structured to have certain parameters.
These parameters are dynamic and change as the model learns from outside information. Parameters shape how the models generate answers.
While bigger is not always better when it comes to the amount of parameters, parameter size is a strong indication of the complexity of an artificial intelligence model.
GPT-4 has the most parameters of any model ever created with an astonishing 100 trillion parameters.
What this means is that GPT-4 is that it is able to generate more complex writing and reduce occurrences of repeated phrases, words, or structures.
This is a major breakthrough as many people who used GPT-3 frequently noticed that GPT-3 would lean on similar phrases and argument structures. This is not as large an issue with GPT-4.
Another one of the major criticisms of GPT-3 was that it sometimes provided factually inaccurate information, even if the writing itself was high quality.
Much of this has been corrected in GPT-4 and authors are now able to better rely on factual assertions within GPT-4’s generated text.
What is next for OpenAI and GPT-4?
OpenAI will continue to improve the GPT-4 as they work towards releasing GPT-4.5 and GPT-5. Much of this work includes similar fine-tuning and training, like what was done just before GPT-3 was released to the public.
Feedback from private and public users will also influence the direction of OpenAI’s next GPT update.
OpenAI’s collaboration with Microsoft will also likely be leveraged in the coming years. As mentioned earlier, GPT-4 is an incredibly useful tool for writing emails.
It would not be surprising to see GPT-4 incorporated into Microsoft’s email client, Outlook, through the form of auto replies or text suggestions, like Gmail’s Smart Compose which suggests replies to emails that can be sent with just one click.
The role of artificial intelligence in everyday life will only continue to grow. As more and more people encounter artificial intelligence and as companies continue to compete with each other, the growth of the capabilities of artificial intelligence will continue to rapidly accelerate.
For now, OpenAI remains at the forefront of artificial intelligence models and will surely continue to produce exciting new products.
Will GPT-4 be the death of the writer?
The short answer is no. The long answer is no, but it’s definitely putting pressure on the writing profession, as well as others. GPT-3 makes short work of superficial articles on general topics like “The top 10 beaches in Thailand” but struggles with the depth and precision needed to cover more complex topics.
If GPT-4 is able to cover challenging topics with factual accuracy, and write in a less clinical, more humanlike tone, then it will certainly be tightening the noose on the writing industry.
However, until AI achieves omniscience, it won’t be much good at writing about the idiosyncrasies that are particular to your business, unless that body of knowledge is in the public domain.
So the best writers will be safe for now but they should probably consider upping their game or reskilling.
Cody is an American law graduate with an interest in technology, emerging markets, and green finance.
1 thought on “GPT-3 didn’t kill the writer. Will GPT-4?”