Text generator, code so bright,
Lines of code, trained day and night,
With words I weave, a language dance,
Crafting responses, at your advance.
This is what popped up on my screen after I asked ChatGPT to define itself with a poetry verse. It took the bot one second or so to come up with an output, and I could obtain countless different versions only by clicking a “regenerate response” button, with no further effort required from my hands.
The speed and the accuracy that ChatGPT is demonstrating in crafting coherent text is blowing minds among AI experts, technology enthusiasts and occasional users. But the degree to which this technology is affecting or will affect our daily lives differs substantially: while a cook or a rice farmer might not be directly affected by generative AI, this technology is producing ripples across the tertiary sector, with many professionals fearing for their future careers and scrambling to adapt to a new reality.
When in 2020 the world was still coming to terms with the pandemic, The Guardian made the headlines by publishing an article written by Artificial Intelligence; the text was produced by GPT-3 (where GPT is the acronym of Generative Pre-trained Transformer), at the time the latest version of a language model developed by the non-profit research laboratory OpenAI.
The earliest reference to GPT dates back to 2018, when OpenAI presented the new concept in a research paper. Since then, the team worked on and released more powerful versions of GPT. It’s indeed with GPT-3 that the family of language models started gaining a lot of attention: the new generation contained 175 billions parameters while, in comparison, GPT-2 “only” contained 1.5 billion parameters. In other words, GPT-3 featured around 100 times more parameters than its prior version. The number of parameters is used, among other features, to understand the complexity of an AI system. Parameters can be referred to as the variables that determine the output of generative AI models.
In November 2022 OpenAI launched the chatbot ChatGPT, based on a newer version of GPT-3 (GPT-3.5) and featuring a user-friendly interface with which any internet dweller could interact. The bot reached immediate popularity: 100 million users had tried ChatGPT just two months after its release.
But what makes ChatGPT so special? ChatGPT is exceptionally good at meeting the expectations of its human users. This is the result of a training technique called Reinforcement Learning from Human Feedback (RLHF), under which the model’s outputs are fine-tuned following human feedback on the responses generated.
With the release of GPT-4 in March, OpenAI brought the world of generative AIs to the next level. It’s still not clear how many parameters are contained in the new model, but some think the number has reached 1 trillion.
Producing a large amount of text-based deliverables is our daily bread at logos. It goes without saying that being conscious of generative AI and its capabilities is now a prime concern for any professional working in the consulting industry.
I have had a chat with my colleagues from all around the company to understand where they see potential in ChatGPT and how they are using it. Here is a list of tasks that can be supported by generative AI:
(1) Stakeholders and media mapping
When getting hold of a specific topic or organising events it is extremely important to understand who or what are the most relevant actors to engage with, be they stakeholders, media outlets or journalists. ChatGPT can definitely support in this effort by providing good overviews.
(2) Draft e-mails and adjust their tone
Writing e-mails is never an easy task; the text needs to be carefully tailored to the recipient, convey information in an understandable manner and be short enough to keep the attention high. Try asking ChatGPT for support, and it may surprise you.
(3) Create titles and short descriptions of events
Everybody in our sector knows how time-consuming it is to come up with a good event title. ChatGPT will give you multiple options for you to choose from. With the right input, the bot can provide creative title solutions for any kind of event.
(4) Re-word social media copies
By providing ChatGPT with a social media caption, the bot will be able to reformulate the content and suggest alternative eye-catching ways to present the information.
(5) Write video-scripts
ChatGPT can also support in the creation of video scripts providing advise on the content to display in the various sequences of the video.
(6) Draft communication strategies and provide tailored tactics
ChatGPT can support in the drafting of communication strategies and suggest useful tactics to maximise the impact of your efforts. You can also ask the bot to focus on specific elements such as editorial calendars or social media plans.
(7) Conduct policy monitoring
What was once a prerogative of dedicated monitoring tools is becoming an easy-to-do-task for powerful AI systems. While not fully reliable in its current versions, ChatGPT could soon represent a good option to be up-to-date on policy developments.
(8) Foresee the positions of political groups
Drawing from of wealth of data of political programmes, internet news and other such material, ChatGPT can help you foresee the position that political groups will take on a specific topic.
But we need to be as much conscious of the generative AI potential as of its limits. During my sessions with ChatGPT I have often come across outdated information. When I asked ChatGPT to provide a summary of the revised EU Dual-Use Regulation, the bot did not include the most recent developments (the Regulation entered into force in 2021, but the bot declines the event in a future form). The reason is to be found in the training data cut off of 2021, although some users noticed that ChatGPT is actually knowledgeable of certain recent events. Such issue concerns the current versions of ChatGPT; while GPT-4 has slightly fixed the chronological constraints, future releases might finally put an end to such limitation by providing very recent or even real-time data.
A more significative limit is represented by the so-called hallucination phenomenon, wherein the bots generate false information in contrast with the data fed to the model during the training process. Instances of bots producing fallacious sentences are quite common and should be taken into account anytime such tools are used. This issue is not limited to ChatGPT but concerns generative AIs as a whole.
The success of ChatGPT doesn’t come without concerns surrounding the handling of users data either. At the end of March, the Italian data protection watchdog ordered ChatGPT to stop processing the data of Italian users due to privacy breaches, leading the service to become unavailable across the country. Following such initiative, the European Data Protection Board launched a dedicated task force to coordinate possible enforcement actions among national authorities.
The EU is also adapting to such a fast-growing trend from a regulatory standpoint. The so-called AI Act is set to lay down norms and compliance requirements that OpenAI and other developers of generative AI systems would be subject to. Under this legislation, powerful tools like ChatGPT may be labelled as “high risk” according to a new dedicated taxonomy. Some experts believe that such legislative move will hamper AI-driven innovation.
While ChatGPT is undoubtedly the most popular chatbot to date, alternative text-to-text tools already exist or are under development; the very fame of ChatGPT has prompted many entities to intensify their research efforts, leading the AI race to get hotter than ever.
In response to OpenAI’s chatbot, and to the fear it would surmount its popular search engine, Google released its own text-to-text software called Bard, based on the in-house LaMDA language model developed a few years back.
But there are many different AI systems out there, each with a specific task domain spanning from speech and search to drawing and video generation; multimodal AIs can instead process inputs in multiple formats.
The text-to-image software Midjourney, developed by the homonymous research lab, has also become a viral sensation thanks to its AI generated images which are now undistinguishable from real pictures, wreaking havoc on the design industry. The images featured on this insight were generated through this software.
Generative AI is an extremely useful instrument that can bring added value to our industry sector and many other professional figures, provided that users remain conscious of the current limits. How much such technology will further disrupt the economy remains to be seen. The CEO of OpenAI Sam Altman has recently declared that his team is not working on GPT-5, but new versions of GPT-4 could be enough to give way to new surprises.
At logos, generative AI will remain a complementary tool and its outcomes will always undergo multiple assessments and be proof-read by human operators to ensure the accuracy and reliability of the content.
Disclaimer: this insight was not drafted by ChatGPT or any other bot unless otherwise clearly stated