Table of Contents
Much has been said about the launch of ChatGPT, the prototype AI-powered chatbot that can have a conversation with users through a simple and intuitive interface. The software created by OpenAI has been welcomed with curiosity and enthusiasm by techies and non-techies alike, fascinated by its potential and the impact it could have on the world of business and society in general.
But there has also been criticism and alarm from those who see a potential threat for some industries, for copyright and for privacy. Indeed, trials conducted to date have highlighted a series of errors and risks which we’ll talk about at the end of the article.
But first, let’s see what ChatGPT is and how it works, what it can do and how it can be used in e-commerce to automate tasks and improve the user experience. We’ll then examine the current limitations of the software, what’s new in ChatGPT 4 and the features of Bard, Google’s rival conversational artificial intelligence system.
What is ChatGPT?
ChatGPT is a prototype chatbot with conversational artificial intelligence, still undergoing testing by OpenAI, the non-profit-making research organisation that developed it and launched it in November 2022. ChatGPT is an updated version of GPT-3.5 (an acronym for Generative Pre-Trained Transformer 3.5), a language processing model based on deep learning also developed by OpenAI.
Unlike the other Instruct models, however, ChatGPR was optimised using a technique known as ‘reinforcement learning from human feedback‘: various human instructors trained the software by providing both the questions and desired answers to various questions.
Drawing on a database filled with a huge number of texts taken from the web, the chatbot comes up with detailed and relevant replies to the user’s questions in a few seconds, using fluent and natural-sounding language. Based on the input it receives, the software can create text similar to that written by a human in various areas: from writing original content and translating texts to generating code in various programming languages.
How it works and what it can do
All you need to do to use ChatGPT is register free of charge on the OpenAI website and, once you have finished this procedure, type your request into the interface on the platform. If the reply you get is inaccurate or incomplete, you can interact with the software by reformulating your question, as the software is able to take the messages you exchange into consideration and gradually refine the content it offers. Users also have the option to click on the ‘Regenerate response‘ button to get a different answer to the one given originally.
There is also a paid plan costing $20/month, which offers access to the tool even during peak times (when the system is slower or even inaccessible), quicker results and the chance to preview new functions.
As we mentioned above, the chatbot can perform numerous tasks, providing a response both to generic requests (‘Invent a recipe containing courgettes and tofu’, for instance) and more specific, technical questions, like writing a article or a story. Here are some examples of things ChatGPT can do:
- Writing text – Product descriptions, essays, recipes, poems, article, video scripts, advertising slogans: the chatbot’s creativity is wide-ranging and versatile.
- Sumarising texts and acting as a search engine – ChatGPT can summarise long conversations and complex documents perfectly. It can also provide a descriptive text on a potentially infinite list of topics, doing the work of a search engine.
- Machine translation – The model can translate the text input into various languages with a decent level of accuracy.
- Generating code – The software can be used to write code in various programming languages (HTML, CSS, Python and C++) and to help users develop software and apps.
How to use it to improve your online shop
As we have seen, ChatGPT is a very powerful and versatile tool, with skills that range from copywriting and software development to translation and marketing. Its AI can even alter the lexicon, grammar and syntax of a text based on the parameters defined by the user. It can write a poem, a story in the style of Star Trek or a letter with a child’s simple vocabulary in a few seconds.
Below is a by no means an exhaustive list of what ChatGPT can be used for in an e-commerce business:
- Suggestions for topics – Are you short of ideas and in need of a few pointers for contents for your blog, newsletters, social media posts or podcast? Ask ChatGPT to give you a list of original and interesting topics for your target audience.
- Proofreading – The software quickly detects grammar mistakes and typos in text.
- SEO – If you provide a list of keywords and a few drafting guidelines, you can get a complete SEO text in a few seconds.
- Content creation in multiple languages – As we noted above, the chatbot can translate large quantities of text quickly into various languages, an extremely useful function for those who sell their products overseas and need a multilingual website.
- Content writing – ChatGPT can generate texts of various types at surprising speed, including blog articles, interview questions, social media posts, product data sheets, promotional emails and adverts. Depending on how well written and relevant they are, you could either use them directly or rework them to your heart’s content.
- Customer service – Using OpenAI’s API, it’s possible to build the chatbot into your e-commerce platform to make customer service more efficient and automate many tasks. For example, ChatGPT can automatically answer customer questions about products and services, place an online order at a customer’s request and send a purchase confirmation email. What’s more, it can also manage queries about order changes, cancellations or returns.
Finally, similarly to other AI software, it can recommend products and special offers to customers based on their profile and previous purchases, using upselling and cross-selling techniques. In his article AI marketing: how artificial intelligence helps businesses to sell, Alberto Maestri provides some interesting examples of how using AI can lead to more personalised marketing strategies.
In summary, e-commerce businesses can use conversational AI to:
- Generate more effective, original and SEO-friendly content quickly and in multiple languages;
- Automate certain operations in the customer service and marketing departments;
- Personalise marketing strategies in line with the characteristics and preferences of each customer segment.
Automating repetitive tasks and customising content brings a range of benefits: an improvement in the user experience, greater customer engagement and loyalty and, with these, more visits to the site and an increase in the conversion rate.
Current limitations and ethical considerations
As OpenAI freely admits, ChatGPT is far from a perfect tool. Sometimes it can provide incorrect, inaccurate or completely false information (so-called “hallucinations”). So it’s important to always check the system’s output to avoid unwittingly spreading disinformation.
It is also worth remembering that the database it was trained on is stuck in 2021, so it cannot provide accurate answers regarding more recent events. It is also immediately obvious that the software tends to be verbose and repetitive, due to certain biases in the training data and over-optimisation issues.
Another risk is the generation of dangerous instructions, violent speech or descriptions that reflect gender or racial prejudices. The moderation API can be used to report or block certain types of unsafe content, but can still give false positives and negatives. To address the problem, OpenAI has recently hired an army of data labellers tasked with teaching the model how to spot and weed out inappropriate content from answers.
Finally, a piece of research by the cyber security company WithSecure showed that ChatGPT can be used to generate texts that can be used for phishing operations, fake news, malware and other dangerous content. The model clearly doesn’t have the sensitivity and emotional intelligence of a human being and cannot make decisions based on ethics or common sense.
The upshot of all this is that while ChatGPT can outperform other software, it cannot replace human involvement entirely. If you run an online shop, you’ll have to assess whether the chatbot can support your business by increasing its efficiency and productivity, but without forgetting that no software can yet match human intelligence and creativity.
ChatGPT 4: what’s new compared to GPT 3.5
On 14 March, OpenAI officially unveiled GPT 4, the new version of the generative artificial technology that powers ChatGPT. As well as being more secure and reliable than its predecessor, the model is multimodal, in other words, it can also create images, graphics and screenshots of various types, as well as text-based responses.
Furthermore, GPT-4 can handle much larger text inputs – up to 25,000 words at once – to help the user in a range of creative and technical writing activities. However, OpenAI also said that the latest version is not immune to the limitations posed by social biases, hallucinations and malicious inputs. GPT-4 is currently only available to ChatGPT Plus subscribers and, in the form of an API, to developers.
Bard, Google’s AI chatbot
As predicted, at the end of February Google announced the imminent launch of the beta version of its conversational artificial intelligence system, Bard. The news came just as Microsoft said that it wanted to build ChatGPT functionality into its Bing search engine.
According to Google CEO Sundar Pichai, Bard will be gradually integrated into Google Search and will directly use indexed content to provide information that is always of high quality and up to date.
Google’s experimental conversational AI service is powered by the new version of LaMDA (Language Model for Dialogue Applications), which was developed a few years ago by the company. But how does it compare to ChatGPT?
The first significant difference, as we’ve seen, is that Bard draws information from the web, whereas ChatGPT was trained on data that was up to date at the end of 2021. Furthermore, according to Google, its AI software will be able to answer complex queries that do not have a single answer by offering a summary of different points of view on the topic. As for similarities, both are conversational artificial intelligence models that can be built into websites, apps and chatbots to improve their functionality and automate certain tasks.
To wrap up, it seems like the conversational AI race between the tech giants will lead to a gradual but significant shift in how we work, look for information and interact with machines and other human beings, opening up fascinating new opportunities.