ChatGPT: Here’s what you need to know
The release of OpenAI's ChatGPT has triggered a new global race in AI. So what is it, how are businesses using it, and what impact could it have on corporate communications? Audley answers your questions.
The launch of ChatGPT in November 2022 captured the attention of businesses and the public in dramatic fashion. Its advent has enlivened debate about the merits and demerits of using artificial intelligence (AI) and what effect it may have on society. Meanwhile, businesses have begun deploying ChatGPT in their products and service offerings.
The newest iteration, ChatGPT-4, does have remarkable abilities and shows genuine utility for business – it is more than a simple novelty. This has led to increased excitement about the promise and potential of AI. However, GPT-4 is not without its flaws and demonstrates that we are still some way from having reliable and truly transformational AI tools at our fingertips.
Naturally, many of Audley’s clients and businesses in our network have taken a keen interest in ChatGPT. The Audley team has started to test the usefulness of the tool for our work and services that we give to clients.
Audley has identified a few useful ways that ChatGPT-4 can be deployed in the context of external communications activities, which are listed below. The Audley view is that GPT-4 can save time with certain tasks, but due to its limitations it is for now an interesting but not particularly valuable tool.
This short briefing looks into how ChatGPT-4 is being used by businesses across the world, explores what the tool can and can’t do well, and why it is provoking debate.
WHAT IS CHATGPT?
ChatGPT is a computer language model that interacts with users and leaves the impression that you, the user, are communicating with a person rather than a machine. The tool is capable of understanding intent. GPT-4 is the fourth iteration of the Generative Pre-training Transformer language model developed by OpenAI, an artificial intelligence research lab based in San Francisco.
This latest version is perceived as a big leap in the field of natural language processing (NLP). It can understand prompts and generate text that would otherwise appear to have been written by a human. More generally, it is thought to have the potential to revolutionise our interaction with machines.
WHAT IS THE ULTIMATE HOPE FOR CHATGPT?
The ultimate aim of GPT and similar language models is to advance the field of NLP and AI to a point where they can understand prompts and generate human-like text at a level that is indistinguishable from human-written text. This would enable a wide range of applications, including but not limited to:
Content creation: GPT could generate high-quality text for various purposes such as news articles, marketing copy, creative writing, and social media posts, which could streamline content creation processes and augment human writers.
Conversational AI: GPT could power advanced chatbots and virtual assistants that can engage in meaningful and contextually relevant conversations with users, providing personalized assistance, answering questions, and performing tasks.
Language translation: GPT could improve machine translation systems by generating more fluent and accurate translations across different languages, bridging language barriers and facilitating global communication.
Education: GPT could be used as an interactive learning tool, providing explanations, answering questions, and offering personalized feedback to learners in various domains, ranging from science and maths to literature and history.
Research and knowledge discovery: GPT could help researchers analyse large amounts of text data, extract insights, and generate summaries or abstracts of research articles, accelerating the pace of scientific discovery.
How is ChatGPT-4 currently being used by businesses?
Duolingo, the language education app, has launched two new features using GPT-4 technology: “Roleplay”, which allows user to practice conversations in the language they are learning with an AI character, and “Explain my answer” which delivers the user feedback on why their answer in a learning exercise was right or wrong.
Stripe, a digital payments platform, is using GPT-4 to enhance its products and experiences for users, e.g., helping users manage fraud and increasing conversion rates. One particular use will be allowing Stripe's customers to make queries about their business analytics using natural language instead of needing to write database queries.
Klarna, a financial technology company providing ‘buy now, pay later’ products, is developing a service using GPT-4 to offer personalised shopping recommendations. This appears to be Klarna’s next step in its effective aim to become the ‘search engine for shopping’ or ‘single shopping destination’ allowing users to find products and compare prices across many different retailers.
Expedia, the online travel agency, is deploying GPT-4 to give users advice on where to go, where to stay, what to see, and what to do on a holiday. Expedia’s CEO, Peter Kern, said the aim is to make the discovery experience easier for people by allowing users to make natural language queries.
OpenTable, the restaurant booking tool, is collaborating with GPT-4 to provide restaurant recommendations. The idea is to allow the tool to generate responses to questions such as: “Where’s a highly reviewed place suitable for a business lunch within a 10 minute walk of Trafalgar Square?”
POTENTIAL USES OF GPT-4 FOR COMMUNICATIONS AND CAMPAIGNS
Representatives of the corporate communications and campaigns industry have begun to discuss how GPT-4 could be used practically:
Content creation
Creating a first draft of a press release or speech, when provided with summary text or bullet points for what the narrative should include – but Audley finds what ChatGPT-4 generates is typically uncreative, generalised, or clichéd.
Generating ideas for headlines for articles and social media posts when provided with the text of the story.
Repurposing articles or similar long-form content by creating different versions based on the original text.
SEO and digital content
Creating different versions of text to populate digital channels with SEO-optimised biographies or company information
General tasks
Generating summaries of large amounts of text and lengthy documents.
Creating transcript of videos.
WHAT ARE THE LIMITATIONS OF CHATGPT-4?
First and foremost, ChatGPT-4 is not fully reliable. It “hallucinates” facts and makes reasoning errors. This means care should be taken when using its outputs.
ChatGPT-4 cannot provide real-time information. It draws upon past online information and it is not connected to the internet.
It cannot be used for research of events that have taken place after September 2021. ChatGPT-4 currently has a cut-off date of September 2021. This means that researching or seeking information and recommendations on anything past that date might be excluded from any findings delivered by the tool.
It often provides inaccurate information on people. When looking for information on individuals it can confuse information related to people carrying identical names or even members of the same family carrying the same surname.
It is not useful for SEO analysis. ChatGPT-4 doesn’t give specific advice on what is required to improve a page’s SEO. It only provides general guidance.
It will only accept text as an input. You cannot upload video or audio.
It will only accept a limited number of characters. If you need to input a longer text stream for analysis, translation, etc., you cannot do it all at once.
WHAT ABOUT GOOGLE BARD?
ChatGPT has some emerging competition. Bard has been described as Google’s ‘answer’ to ChatGPT with the aim of keeping Google as the market leader in internet search.
There are obvious similarities: ChatGPT and Google Bard are both natural language AI chatbots, which in practice means they offer similar solutions and work the same way. They are both designed to be asked questions and then provide answers.
The two tools are trained on different data models, and Google Bard allows users to give prompts via voice using their device’s microphone. The tool also includes a “Google it” button which provides users links to continue research their outside of Bard.
Google Bard suffers from similar drawbacks to ChatGPT, including inaccuracies in its responses. Bard has only been available since March 2023 and a lengthy waitlist makes it hard to gain access to. For the moment, the prevailing industry view is that Bard is a ‘decent alternative’ to ChatGPT.
WHY IS CHATGPT-4 PROVING CONTROVERSIAL?
ChatGPT-4 and other NLP models are showing unprecedented capabilities in generating human-like text and have the potential to disrupt certain industries. The speed of ChatGPT’s evolution has intensified debate and increased concerns about the ethical implications of AI-generated content, such as issues related to bias, authenticity, and potential misuse.
In March, influential technology industry leaders and academics signed an open letter calling on the development of AI tools like Chat GPT-4 to be temporarily paused to allow time for AI developers and policymakers to accelerate the creation of AI governance systems. The signatories said that AI labs were now “locked in an out-of-control race to develop and deploy ever more powerful digital minds that no one – not even their creators – can understand, predict, or reliably control.” They argued that “powerful AI systems should be developed only once we are confident that their effects will be positive and their risks will be manageable.” Co-signatories included Elon Musk, Steve Wozniak and Yuval Noah Harari.
In April, the Italian government moved to ban Chat GPT-4. The Italian data regulator, Garante, believes that GPT-4 breaches General Data Protection Regulations (GDPR) and will be investigating it accordingly. The regulator argued there was no legal basis to justify “the mass collection and storage of personal data for the purpose of 'training' the algorithms underlying the operation of the platform.” This move has prompted other European regulators to see if they should follow suit. Dessislava Savova, a partner at Clifford Chance, said that the move of the Italian regulator shows that “GDPR does offer tools for the regulators to be involved and engaged into shaping the future of AI.”
Recommended reading
Battle of the chatbots: how does GPT-4 stack up against Bard?, Financial Times, Madhumita Murgia and Kari-Ruth Pedersen, 3 April 2023.
Your Humanity Could Save Your Writing Job from ChatGPT-4, Bloomberg, Adrian Woolridge, 21 March 2023.
Italy's ChatGPT ban attracts EU privacy regulators, Reuters, Supantha Mukherjee, Elvira Pollina and Rachel More, 3 April 2023.
GPT-4 Will Make ChatGPT Smarter but Won't Fix Its Flaws, Wired, Will Knight, 14 March 2023.
AI Doesn’t Hallucinate. It Makes Things Up, Bloomberg, Rachel Metz, 3 April 2023.
Review: We Put ChatGPT, Bing Chat, and Bard to the Test, Wired, Lauren Goode, 30 March 2023.
By Rolf Merchant, Director, Audley