The biggest topic in tech right now, is undoubtedly ChatGPT-3 and how it could potentially re-shape certain industries.
But where did Chat-GPT come from? Why is it so dominant? And how is it currently being used to give some, an edge? Lets take a look..
AI bots, transform and roll out
I’m sure you’re all familiar with Generative Pre-Trained Transformers. In essence, they are state-of-the-art language processing models researched and developed by OpenAI, with their first paper on the topic being released in June 2018.
These models quickly gained attention within the academic world for their ability to generate human-like text and answer queries, and there has been various releases of the GPT models over the last four years. Late 2022 brought the latest incarnation of the model, Chat-GPT3, and saw the tool’s popularity explode. In a matter of weeks, it has become entirely mainstream and globally used. The velocity of ChatGPT3’s user acquisition enabled it to rack up an astonishing one million users within a week. For context, it took Twitter 2 years to achieve the same feat and even Facebook needed 10 months to reach that milestone.
Prior to ChatGPT-3, there was the development of GPT-1 and GPT-2, which were released in 2018 and 2019 respectively. These models were trained on a large corpus of text and data and were able to generate human-like outputs by themselves. However, the models’ capabilities were limited, with additional tuning needed to help them perform specific tasks. The first two iterations of the model also had a tendency to generate text that was biased or factually incorrect.
A major step forward
So, how exactly did ChatGPT-3 improve on its predecessors to such a degree that it effectively took the tech world by storm and turned its developer, OpenAI, into a start-up valued at a cool $29 billion? Well, the main difference between ChatGPT-2 and ChatGPT-3 is quite simply the size of the model, with the former being trained on 1.5 billion parameters and the latter a dwarfing 175 billion, making it 116 times bigger! The increase in size helps the model be better at what it does, by effectively making it more accurate when predicting the next word in a sentence, meaning it is better at question answering, summarisation, and natural language understanding.
The outputs of ChatGPT-3 are consequently far more coherent and fluent, even when only posed with a small amount of context as a question. Its ability to maintain a conversation by adapting to follow up questions from its answers, is on the surface very impressive. The fact it is trained on over 8 million web pages, books and factual text sources, means it retains significant knowledge and the ability to answer complex queries instantly. This has caused ethical concerns that it could directly challenge some roles and industries with its ability to perform some tasks much quicker than humans, potentially to a better, more consistent standard. But can Chat-GPT3 really threaten jobs?
The answer, somewhat intriguingly, is both yes and no.
The potential impact of ChatGPT is of course immense, but the approach many are taking is to not try and compete with ChatGPT, but to determine how to leverage it so that it can be used to accelerate existing workflows at various stages.
Make the tech work for you
Embedding its capabilities into phases such as keyword research or content topic generation, will very likely help accelerate proof-of-concept and ideation stages. This in turn should allow for more time to be spent on the human refinement of work at critical points of differentiation, allowing for an increasing amount of time for creativity where it is most needed.
The flip side of this scenario, is that professionals who are unwilling to adopt and leverage the capabilities to enhance and quicken appropriate deliverables of a project, will likely find themselves delivering work at a slower rate than those competitors who do.
Whilst the capabilities and applications for ChatGPT-3 are indeed vast, we completed some exploratory work around international money transfer and related topics for basic content ideation.
1. We first picked a topic and requested a list of related entities:
2. And then asked for a topical cluster around the same topic:
3. With no real specific requirement we asked for a content brief on any one of the titles provided:
An impressive start, considering this could be a completely new topic to the user, or at least one with which they are struggling to find direction. At this point, we could choose to take this output and begin to craft our own content piece around the topic.
But we push on just to explore ChatGPT further and test some more of its capabilities.
4. We asked ChatGPT to provide us Q & As from Google PAA, to give us an idea of the topics users are frequently searching answers for. This is a great start for allowing us to shape and structure our proposed content.
You’ll notice that ChatGPT can’t currently access Google, so its results are limited to data from circa. 2021 and are not in real-time. This appears to be one of the few limitations in its current format.
5. A quick comparison of the best providers in this space.
6. Finally, we picked out one of the providers and asked that it create a 2000 word article on using that provider, over that of Cryptocurrency.
7. Finally, thinking of a way in which we may want to wrap-up the above article with instructions around a CTA, in this case, opening an account with the provider, we ask for a quick “how-to” which we could include.
The conversational aspect of ChatGPT-3 means that we could of course have refined the output at any of the stages above, by posing different questions or simply asking follow-ups to probe deeper in to any of the answers. However at this stage, many are still trying to understand how ChatGPT best works for them and are learning how to leverage alongside their own creative touch, so the example we went through is just exploratory.
One thing is for certain, the use cases for ChatGPT-3 are far-reaching and its influence will likely be felt across many industries very soon.